pax_global_header00006660000000000000000000000064136662205170014522gustar00rootroot0000000000000052 comment=b642639117eafedc760d8b84c0d2c4872b0da084 samblaster-v.0.1.26/000077500000000000000000000000001366622051700142115ustar00rootroot00000000000000samblaster-v.0.1.26/LICENSE.txt000066400000000000000000000021001366622051700160250ustar00rootroot00000000000000The MIT License (MIT) Copyright (c) 2013-2020 Gregory G. Faust Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. samblaster-v.0.1.26/Makefile000066400000000000000000000010271366622051700156510ustar00rootroot00000000000000# Determine the samblaster build number BUILDNUM = 26 # INTERNAL = TRUE OBJS = samblaster.o sbhash.o ifdef INTERNAL PROG = samblaster$(BUILDNUM) CCFLAGS = -Wall -Winline -Wliteral-suffix -O0 -g -D BUILDNUM=$(BUILDNUM) else PROG = samblaster CCFLAGS = -Wall -O3 -D BUILDNUM=$(BUILDNUM) endif CC = gcc CPP = g++ CFLAGS = $(CCFLAGS) -std=gnu99 CPPFLAGS = $(CCFLAGS) SAMBLASTER: $(PROG) $(PROG): $(OBJS) $(CPP) $(LDFLAGS) -o $@ $(OBJS) %.o: %.cpp $(CPP) $(CPPFLAGS) -c $< -o $@ clean: rm -f $(PROG) *.o *.co *.linkinfo samblaster-v.0.1.26/README.md000066400000000000000000000272171366622051700155010ustar00rootroot00000000000000# *samblaster* **Written by:** Greg Faust (gf4ea@virginia.edu) [Ira Hall Lab, University of Virginia](http://faculty.virginia.edu/irahall/) **Please cite:** [Faust, G.G. and Hall, I.M., “*SAMBLASTER*: fast duplicate marking and structural variant read extraction,” *Bioinformatics* Sept. 2014; **30**(17): 2503-2505.](http://bioinformatics.oxfordjournals.org/content/30/17/2503) **Also see:** [SAMBLASTER_Supplemental.pdf](https://www.dropbox.com/s/s7vvdtf2gmytvay/SAMBLASTER_Supplemental.pdf?dl=0) for additonal discussion and statistics about the duplicates marked by *samblaster* vs. *Picard* using the NA12878 sample dataset. Click the preceeding link or download the file from this repository. --- **Current version:** 0.1.26 Support for Linux and OSX (Version 10.7 or higher). ## Summary *samblaster* is a fast and flexible program for marking duplicates in __read-id grouped1__ paired-end SAM files. It can also optionally output discordant read pairs and/or split read mappings to separate SAM files, and/or unmapped/clipped reads to a separate FASTQ file. When marking duplicates, *samblaster* will require approximately 20MB of memory per 1M read pairs. ## Installation *samblaster* is self contained and therefore has no installation dependencies beyond **g++** and **make**. *samblaster* can be downloaded from the **_releases_** tab or manually downloaded via *git clone*. Afterwards, simply use *make* and copy *samblaster* to a directory in your path. For example: ~~~~~~~~~~~~~~~~~~ git clone git://github.com/GregoryFaust/samblaster.git cd samblaster make cp samblaster /usr/local/bin/. ~~~~~~~~~~~~~~~~~~ ## Usage See the [SAM File Format Specification](http://samtools.sourceforge.net/SAMv1.pdf) for details about the SAM alignment format. By default, *samblaster* reads SAM input from **stdin** and writes SAM to **stdout**. Input SAM files usually contain paired end data (see [Duplicate Identification](#DupIdentification) below), must contain a sequence header, and must be __read-id grouped1__. By default, the output SAM file will contain all the alignments in the same order as the input, with duplicates marked with SAM FLAG 0x400. The **--removeDups** option will instead remove duplicate alignments from the output file. __1A read-id grouped__ SAM file is one in which all alignments for a read-id (QNAME) are grouped together in adjacent lines. Aligners naturally produce such files. They can also be created by sorting a SAM file by read-id. But as shown below, sorting the input to *samblaster* by read-id is not required if the alignments are already grouped. **COMMON USAGE SCENARIOS:** To take input alignments directly from _bwa mem_ and output to _samtools view_ to compress SAM to BAM: ``` bwa mem samp.r1.fq samp.r2.fq | samblaster | samtools view -Sb - > samp.out.bam ``` When using the *bwa mem* **-M** option, also use the *samblaster* **-M** option: ``` bwa mem -M samp.r1.fq samp.r2.fq | samblaster -M | samtools view -Sb - > samp.out.bam ``` To additionally output discordant read pairs and split read alignments: ``` bwa mem samp.r1.fq samp.r2.fq | samblaster -e -d samp.disc.sam -s samp.split.sam | samtools view -Sb - > samp.out.bam ``` To pull split reads and discordants read pairs from a pre-existing BAM file with duplicates already marked: ``` samtools view -h samp.bam | samblaster -a -e -d samp.disc.sam -s samp.split.sam -o /dev/null ``` To process a BAM file of singleton long reads to pull split and/or unmapped reads with/without marking duplicates: ``` samtools view -h samp.bam | samblaster --ignoreUnmated [-e] --maxReadLength 100000 [-s samp.split.sam] [-u samp.umc.fasta] | samtools view -Sb - > samp.out.bam samtools view -h samp.bam | samblaster --ignoreUnmated -a [-e] [-s samp.split.sam] [-u samp.umc.fasta] -o /dev/null ``` --- **OPTIONS:** Default values enclosed in square brackets [] 
Input/Output Options:
-i --input           FILE Input sam file [stdin].
-o --output          FILE Output sam file for all input alignments [stdout].
-d --discordantFile  FILE Output discordant read pairs to this file. [no discordant file output]
-s --splitterFile    FILE Output split reads to this file abiding by paramaters below. [no splitter file output]
-u --unmappedFile    FILE Output unmapped/clipped reads as FASTQ to this file abiding by parameters below. [no unmapped file output].
                          Requires soft clipping in input file.  Will output FASTQ if QUAL information available, otherwise FASTA.

Other Options:
-a --acceptDupMarks       Accept duplicate marks already in input file instead of looking for duplicates in the input.
-e --excludeDups          Exclude reads marked as duplicates from discordant, splitter, and/or unmapped file.
-r --removeDups           Remove duplicates reads from all output files. (Implies --excludeDups).
   --addMateTags          Add MC and MQ tags to all output paired-end SAM lines.
   --ignoreUnmated        Suppress abort on unmated alignments. Use only when sure input is read-id grouped,
                          and either paired-end alignments have been filtered or the input file contains singleton reads.
                          --ignoreUnmated is not recommended for general use on paired-end data. It disables checks that detect incorrectly sorted input.
-M                        Compatibility mode (details below); both FLAG 0x100 and 0x800 denote supplementary (chimeric). Similar to bwa mem -M option.
   --maxReadLength    INT Maximum allowed length of the SEQ/QUAL string in the input file. [500]
   --maxSplitCount    INT Maximum number of split alignments for a read to be included in splitter file. [2]
   --maxUnmappedBases INT Maximum number of un-aligned bases between two alignments to be included in splitter file. [50]
   --minIndelSize     INT Minimum structural variant feature size for split alignments to be included in splitter file. [50]
   --minNonOverlap    INT Minimum non-overlaping base pairs between two alignments for a read to be included in splitter file. [20]
   --minClipSize      INT Minumum number of bases a mapped read must be clipped to be included in unmapped file. [20]

-h --help                 Print samblaster help to stderr.
-q --quiet                Output fewer statistics.
   --version              Print samblaster version number to stderr.
--- **ALIGNMENT TYPE DEFINITIONS:** Below, we will use the following definitions for alignment types. Starting with *samblaster* release 0.1.22, these definitions are affected by the use of the **-M** option. By default, *samblaster* will use the current definitions of alignment types as specified in the [SAM Specification](http://samtools.sourceforge.net/SAMv1.pdf). Namely, alignments marked with FLAG 0x100 are considered *secondary*, while those marked with FLAG 0x800 are considered *supplementary*. If the **-M** option is specified, alignments marked with either FLAG 0x100 or 0x800 are considered *supplementary*, and no alignments are considered *secondary*. A *primary* alignment is always one that is neither *secondary* nor *supplementary*. Only *primary* and *supplementary* alignments are used to find chimeric (split-read) mappings. The **-M** flag is used for backward compatibility with older SAM/BAM files in which "chimeric" alignments were marked with FLAG 0x100, and should also be used with output from more recent runs of *bwa mem* using its **-M** option. --- **DUPLICATE IDENTIFICATION:** A **duplicate** read pair is defined as a pair that has the same *signature* for each mapped read as a previous read pair in the input SAM file. The *signature* is comprised of the combination of the sequence name, strand, and the reference offset where the 5' end of the read would fall if the read were fully aligned (not clipped) at its 5' end. The 5' aligned reference position is calculated using a combination of the POS field, the strand, and the CIGAR string. This definition of *signature* matches that used by *Picard MarkDuplicates*. 1. For pairs in which both reads are mapped, both signatures must match. 2. For pairs in which only one side is mapped (an "orphan"), the signature of the mapped read must match a previously seen orphan. Starting with *samblaster* 0.1.25, forward and reverse strand orphans/singletons are allowed to be duplicates of each other. In an orphan pair, the unmapped read need not appear in the input file. In addition, mapped non-paired singleton (possible long) read alignments will be treated the same as an orphan pair with a missing unmapped read. 3. No doubly unmapped pair will be marked as a duplicate. 4. Any *secondary* or *supplementary* alignment associated with a duplicate *primary* alignment will also be marked as a duplicate. --- **DISCORDANT READ PAIR IDENTIFICATION:** A **discordant** read pair is one which meets all of the following criteria: 1. Both side of the read pair are mapped (neither FLAG 0x4 or 0x8 is set). 2. The *properly paired* FLAG (0x2) is not set. 3. *Secondary* or *supplementary* alignments are never output as discordant, although a discordant read pair can have such alignments associated with them. 4. Duplicate read pairs that meet the above criteria will be output as discordant unless the **-e** option is used. --- **SPLIT READ IDENTIFICATION:** **Split Read** alignments are derived from a single read when one portion of the read aligns to a different region of the reference genome than another portion of the read. Such pairs of alignments often define a structural variant (SV) breakpoint, and are therefore useful input to SV detection algorithms such as [LUMPY](https://github.com/arq5x/lumpy-sv/). *samblaster* uses the following strategy to identify split reads alignments. 1. Identify reads that have between two and **--maxSplitCount** *primary* and *supplementary* alignments. 2. Sort these alignments by their strand-normalized position along the read. 3. Two alignments are output as splitters if they are adjacent on the read, and meet these criteria: - each covers at least **--minNonOverlap** base pairs of the read that the other does not. - the two alignments map to different reference sequences and/or strands. - the two alignments map to the same sequence and strand, and represent a SV that is at least **--minIndelSize** in length, and have at most **--maxUnmappedBases** of un-aligned base pairs between them. 4. Split read alignments that are part of a duplicate read will be output unless the **-e** option is used. --- **UNMAPPED/CLIPPED READ IDENTIFICATION:** An **unmapped** or **clipped** read is a *primary* alignment that is unaligned over all or part of its length respectively. The lack of a full alignment may be caused by a SV breakpoint that falls within the read. Therefore, *samblaster* will optionally output such reads to a FASTQ file for re-alignment by a tool, such as [YAHA](https://github.com/GregoryFaust/yaha/), geared toward finding split-read mappings. *samblaster* applies the following strategy to identify and output unmapped/clipped reads: 1. An **unmapped** read has the *unmapped read* FLAG set (0x4). 2. A **clipped** read is a mapped read with a CIGAR string that begins or ends with at least **--minClipSize** unaligned bases (CIGAR code S and/or H), and is not from a read that has one or more *supplementary* alignments. 3. In order for *samblaster* to output the entire sequence for clipped reads, the input SAM file must have soft clipped primary alignments. 4. *samblaster* will output unmapped/clipped reads into a FASTQ file if QUAL information is available in the input file, and a FASTA file if not. 5. Unmapped/clipped reads that are part of a duplicate read pair will be output unless the **-e** option is used. samblaster-v.0.1.26/SAMBLASTER_Supplemental.pdf000066400000000000000000003266401366622051700211050ustar00rootroot00000000000000%PDF-1.3 % 4 0 obj << /Length 5 0 R /Filter /FlateDecode >> stream x}Gry.1Ob 4 EW*Bz&}]}>D{fe]_X]U{sՏ qsr|ߪ;^o?˹n|jgr?r9Or|^1؂]e/W6N=r?o"Z8@?߿yt_̈́zhw9se*ڏ~{z'O^xzzďOקOmc1CF7_\0<#HL|asZ[ @Cp | +UBWҳ:\3Y7\\&2(VTOu& EFhEb'öϟ!7B6g޾r?9]߃@ €uXƖ`I_ WR_pM4O߀(+|L[ŲopH; m<.s5^+NuK3.`AZb5uR?>}80Ҷ|}UޢNMؗ ldt|&ХH5$8ԋT{b?SxΐFđWq\| Ac'B: ͘I;8JK`>?#6|oN߻FQCrPJWױ9N}̽BZ/P["$XIފX)~OJ"nDC$5%/A"8_cСZ1=0tMK1+6s "ZVs|C&摘=8'?^ oxG?Fa@( Ou#b.TwmFu,=CN:DCoa Ah~3libb@9y{!#ou/Œ}bz p.gh.ZIiBFzKsNUM"=随_.2ZneG| Apқ2ٜt*+A#nN{8BbB:ԥ +r4e0/Ynh`.}]~a{azJ&R*Btkp:(jS}cy{hcV=Ȯ׏y[\N['`xD}CQY2JVf_5NW>CPvI9z,b䃧:)E 10R)=WDBL,T4M =U%T6 !xX}a2AUjq(k SY|w@#bIH[BYnE3yәVd(HbbNUrǤt)6ghc#ٲ5գGT~i0нv=>r,ePchD.'hePݰ7=t=(CG2/Nu!}m/h|k>C=Q qTy|˖,!yas ۄbY hW?)oq#ƾD ! h"HPB%Gv|?qOwQ71)wިSiTSe#)E6I8i5Ҭ]c"YKvpl W#8!Uc4 o=:p'DX*]1U6KC1 B7Pzo Y BNp1cLч_4 g3p׉EIIY"WDs0$R ՓcJL{L"uRvE!G`bQ_KTpG4M@ 8nXp{4:vd=I(l9#nfC9Z%+Ǘۚ^Y?uxѦ r\C2_(ACVϭQ_<|-A'A U#RGKXè KM+1;Y<%G?=WUłn2KzֈoCyd8]iZץpǡ qr9ó4s^;*$2=M ŏndTXԠcIqѠdleTKsz֋_yC1 ~ 0B ,%+L*̀/ NJÈpZ p( 1*JҽxB^JE\;8 Vvp)/ -Ja(ۑ6d@M}']߻ ;VG֯ GL?Z9g9P'ӎ.^ K`ZN|nP 0$Y썙pVVljWxu6eH(  ?Ip kIoWt xBZ'G X#yĚw8{^gȼhP;Ci-AY:J[BNPUZ>5䠕Em`  ( ׄr޺nOΒ 3R$*HCUTGCSsg^.%\IhbB% ϸ/GG佗hfd7g;n϶dKdS t#7 U5]`$uyrtab dJKQY9e/-G.XCumZEilkl_0MNTu1QBo@ij c G`$r E^9b*"%rT"n|٣TDol#c Y3`) _ ߚ 7<}[d>C٤Jn)=ژ Fxzv2[b+4[9l| Ll.+9`D: KP7̅0ߏc]jW@QDo{]ߨT3퇒Fs%4SgY>/8#/$ \lLj5}A\D+۾hj5: p-:7!TwJ7@{G/u}Phqנ<.Th~Վf_o;cr@̊<7zƜLx#)>7'2a.vѯ'j_0 @/_sGl HN^v-vAYqI"[:^ ދ7H`Df Fk{և*PTG!{h*_^}g 58Ǥӿ 4,K >>)2gYV\,σs bP;=|[r<+J|m+6ӁoXqf۠p\+Ngd{h+t,cn-H=qۖG*ܠmTcl7hB Fo`&x % :S؎""K`)&R^S:81߉"Q>T(K^ |8P"g* oXɇS3d?v0 m GN1V|2X=ILOnBM T&/&ޤw̘FEO$hE y|#Kѐ,y'"!@C(yQ1昊;!`MВ p\A J5W,JTLFiu;̽mTwZɹ;ڡw߭d%G ҅3pvժ]&v+nVx=Eip<6bws { 2Y#*q6BCyqׁ5.\; J䘥dN5*9kjS#êwmJ2x͈6Γ g,*lNYcV vi7dENQoIeR&o:>:Epz<(vn \Cȃle<غh n! =r lU xFl5 >@`^yXxP1ۺ;^@ReUXC8fx΄!ɛy)ʛܓX|=(h'XMߊBCӖlSzKGBzeޟzvJUnqaV~LGl;yfii[rd9F{ˆaZƔx%86)ӂdM36|[?0wֶ/|q[g~>--NUZ8Swz4x%k(bQ (@>Yx583dpHyѵXi@ )n"㱰@ Va!(}.d#=42Q,`P=H3 s˴ȣ춑7C\n&C? j9"V'm<>iΆ^hH!Bt+lK4ф3pU҉ G"G]gGuc,dh"sյSž S2vki{М*g2!E4 7ukJ_F.>*:9řb.j;F~H%%bYGbEg9/ s9;_%s9ݡZړ; Galn h4h=TþO ~9ZTNܛjgi=`;tĒneܼp`ᕛyv|B+Y~0_ _,xki=F#Iy,fDY()~8 d%НlZ_CXj1qcDV1;Peg 7N2oepR |nb"DVGA26Jv Z CE6 uj:ȍnY z;diAl @ *ڼ?ʪf:\7;qD5 ¾e8`LAjPo<&ymш)Ϥ~q҆/vgڱ~X;1(N^kGQ]$Z1^FdU87j' -D Mk`uJך3nSV6.ih65ãb/U=$/-f3ȥY5Zڔ|~nT] }V`7QFt 5(%"Y&0ˬNre KQkAzOf)#룦:iVdw/eˈAr/ [R>&kS' !PpG okT{V4&v,,Du%40/B-)V5*E@K5b *i71EIovQmkH+qg1^%#F"~ 7Q[.R@D0#0*U;=s^"R/1WklH-$#Y/9VڨqtQ`/"6EB(0L!BQՁU 9 o/ xZi1Q5;ΓтM}OqMUDQg^3ΩvLjxtAjۙp2 l [ӥO*%}aL&-b  !&T+kNuT'$\cX;1E t{hCt5D!r+VNEhnEPs'@3QyƅL.ǍQW{FiS6VkL~\2a4b]g[Vr\PsF¯pc@ uuq3~HTx3}֐uxqr ՀEX%S/I39'Jh |Ȕ(xdkD+t }k>1lޅ= nIOe-@, ~.d/5r!f5 x_vgų@EnMHR6TJ:=Ø6AUky9x}t;|J NzAjD5 2{H+i#)]#r,Һ7ήguj<9P7,{G{<6&lHgn쾎GMm6 6-跩T䂽Sӕұ-\z5L]$*bZ15ڦd؞m∁roUàjpdxl<.&{8t UGYQ;iLDvh8BUZ GP.@Kaqk}j(+j5ŗӃ3gd9L0#! •DKe8TéɻMΫyc!z4AavlTMSh7x8!lMQ=;%d5/&T1\I*SU6F'#Ihx[m.dmx*,%NdFrk(oԔ,[*~ᨵ_Ъ8|5Y?4ƴ&AA"Frt6SOP lzé ]eBG5\Ss^5|G}#}z{ endstream endobj 5 0 obj 10993 endobj 2 0 obj << /Type /Page /Parent 3 0 R /Resources 6 0 R /Contents 4 0 R /MediaBox [0 0 612 792] >> endobj 6 0 obj << /ProcSet [ /PDF /Text ] /ColorSpace << /Cs1 7 0 R /Cs2 16 0 R >> /Font << /TT3.1 11 0 R /TT4.1 13 0 R /TT1.0 8 0 R /TT6.1 18 0 R /TT5.1 15 0 R /TT2.0 9 0 R >> >> endobj 19 0 obj << /Length 20 0 R /N 3 /Alternate /DeviceRGB /Filter /FlateDecode >> stream xwTSϽ7" %z ;HQIP&vDF)VdTG"cE b PQDE݌k 5ޚYg}׺PtX4X\XffGD=HƳ.d,P&s"7C$ E6<~&S2)212 "įl+ɘ&Y4Pޚ%ᣌ\%g|eTI(L0_&l2E9r9hxgIbטifSb1+MxL 0oE%YmhYh~S=zU&ϞAYl/$ZUm@O ޜl^ ' lsk.+7oʿ9V;?#I3eE妧KD d9i,UQ h A1vjpԁzN6p\W p G@ K0ށiABZyCAP8C@&*CP=#t] 4}a ٰ;GDxJ>,_“@FXDBX$!k"EHqaYbVabJ0՘cVL6f3bձX'?v 6-V``[a;p~\2n5׌ &x*sb|! ߏƿ' Zk! $l$T4QOt"y\b)AI&NI$R$)TIj"]&=&!:dGrY@^O$ _%?P(&OJEBN9J@y@yCR nXZOD}J}/G3ɭk{%Oחw_.'_!JQ@SVF=IEbbbb5Q%O@%!BӥyҸM:e0G7ӓ e%e[(R0`3R46i^)*n*|"fLUo՝mO0j&jajj.ϧwϝ_4갺zj=U45nɚ4ǴhZ ZZ^0Tf%9->ݫ=cXgN].[7A\SwBOK/X/_Q>QG[ `Aaac#*Z;8cq>[&IIMST`ϴ kh&45ǢYYF֠9<|y+ =X_,,S-,Y)YXmĚk]c}džjcΦ浭-v};]N"&1=xtv(}'{'IߝY) Σ -rqr.d._xpUەZM׍vm=+KGǔ ^WWbj>:>>>v}/avO8 FV> 2 u/_$\BCv< 5 ]s.,4&yUx~xw-bEDCĻHGKwFGEGME{EEKX,YFZ ={$vrK .3\rϮ_Yq*©L_wד+]eD]cIIIOAu_䩔)3ѩiB%a+]3='/40CiU@ёL(sYfLH$%Y jgGeQn~5f5wugv5k֮\۹Nw]m mHFˍenQQ`hBBQ-[lllfjۗ"^bO%ܒY}WwvwXbY^Ю]WVa[q`id2JjGէ{׿m>PkAma꺿g_DHGGu;776ƱqoC{P38!9 ҝˁ^r۽Ug9];}}_~imp㭎}]/}.{^=}^?z8hc' O*?f`ϳgC/Oϩ+FFGGόzˌㅿ)ѫ~wgbk?Jި9mdwi獵ޫ?cǑOO?w| x&mf endstream endobj 20 0 obj 2612 endobj 7 0 obj [ /ICCBased 19 0 R ] endobj 21 0 obj << /Length 22 0 R /N 3 /Alternate /DeviceRGB /Filter /FlateDecode >> stream xUoT>oR? XGůUS[IJ*$:7鶪O{7@Hkk?<kktq݋m6nƶد-mR;`zv x#=\% oYRڱ#&?>ҹЪn_;j;$}*}+(}'}/LtY"$].9｟%{_a݊]hk5'SN{<_ t jM{-4%TńtY۟R6#v\喊x:'HO3^&0::m,L%3:qVE t]~Iv6Wٯ) |ʸ2]G4(6w‹$"AEv m[D;Vh[}چN|3HS:KtxU'D;77;_"e?Yqx endstream endobj 22 0 obj 1047 endobj 16 0 obj [ /ICCBased 21 0 R ] endobj 24 0 obj << /Length 25 0 R /Filter /FlateDecode >> stream x}~xg]lS,YLGDXΆcgcirU=TLDwOLd&Pykmt:nø=]mn_~v:]i|e; 틷o~l^os[ Gm8_ tiiCy z۷< mJ8wtdݮO>e]oczn^o3z]]os?osǟ/J UzRs27rkwgeIPmwڍMUpVyWe{j{+v71%H= *FO?&o&G.n#n6|#MߢO_ݭՇ U SZȚGB L>`CV"(-ʧ^ qcD%T:w(BB*%XofgrD`̈́>0z) HFT\4q0: 2FZ_\o(B+|v$6joٿnk~98s@IǖQ[VZ[MhW}ڋo}wl,knˆ6Hj{:to3v8HoŰqyө{б-{}sS݌=gRmcd@ND# ,tM)!íZSkP"Sl9~C :^Aii+Δ(zQթ4?2IszQoi0~ڮ O2-*֛pEʏDiSQ\Kau2IsAmRpjѮ)tYw߅yk?F,v$-W' ⓱A'O:zʡv}3M9ӱK r_o[r#[iR*Izw GD(F/z}~sRN]YCnSbg%lr@; Rg02;zwBm,lEGha »vvf#$a7`o|."diɶ@a/ 5l nɠ%1Co8bAe;""X>IZa;f4Cy [6b=n eh)Bs ]3g#, B"|~vՏT([. ~_i6W+ l`| 6r^ؐc( јUE4sUo PLI@Jx{9ď) SwPǘoNDPߙo(4bitgMbɔ|)6 \0Q߽3Pi#J}J-́3ʍ[thX65hxN: bdxD>!!t3OY9aLAnݼ,?Hybt%}mO)ſqX7+: %ժN ')fƏ"/v|[&7L?e; l6>!.}?:'z gA-+nY㛚L%pG5VԢmojQk..yq g W3sCŬ gY+Q.LAnFGEJYq\TӢV"OK$~eC^l-@ݮ|9Ia7W`RsͥVYY٠SC g(wY-Z: Ji!|4۲Ϻ5!bkxO6f21!(cĦ>F"ƈiwKuL4%w"g"g$^mlxmlZZT3XۮncQu}D>o5fڰV]+T0cо3a,Fk(Tvە `lD2u#fT !">:øhSY81yS 9Y~r=xܦyߘ2 ckDLwI|nèK&6W%-zIq"&h6ld$ Zd zb7X¬nӪVc.Ze`tv\k&]*+vY9X~S\D)$ +TYcSH/@5 cįXzuQǔ<8vURjf(fQQ1V~jgtfdހ@h\ynT3NQ#l=`Yk.F;h]Ke72ˢ:+pF*hen6J*\ lD;2| GZc3ヵ! ѲYt)3(vxey9g@~n`B[kt<-_tL{Դwv=f~qaړھPcd8{&wkS3A$6+'᙮ N S/[2M ~Nd&68 a621Օ , y0lZ yC;ysmbհ03z`&dvh}C.+r'"txF.Yg!΍8 ŮqC)Q^^V3̘b<} [B!Xq1j=([Z:7 }SkxV84*ake>)$^քɴ`5edZ7){!5+%-Ы{Jė#qa |xznGzZUmoNRebk)r 0Rr0]%Rw.뺴~szXڒn?wQdHͲ;#Y8e^;@fm0qa]|߾7ͣ'6t8\O!”Ðpf GHXhVpўr*wۏ ͹8ӄj.|!ϔ kŠ+_،IM{>)[ra'1&U k}6:\=`X54AX͓Iv\Aa6c#+UMz䩦~+mvQ`3Hez6^C~\5]^-Z.J<ӧ",1veM| %yVi)\QЄ%Q1-:Z9q|qw@o19QPo/wq񈳕Hn:wZ駘 $ި6S==qo"Ckg9w8UoHGHN&[qț4wANW A6R=i4ղ'Hqf5-vuB҆!wafS5qQeŵG;-',YX P?&sgCߔH~O$#Y̆s%C˜t4fE7Ւ4TCܐ2M2*h˹wyn1:֤)'i7{[\03O>S#S&+ٌ  WrtNðejnq DtO4.ՊZ(+mqpڮқ+#U-~c$B)n1`QIXm̻ŪZW ןQL{2zpϷ`xYeS+gQո& h|u(|>)F׬sx}fg vx;^.(CďϫmqKm0mOjءsC0ȱ&z\Xr,5"EvWUgcڡEH4.e4dhSFfkz~Kmx&b6%||w&ߚGcDw(sQi$aF% ʷNGe2w :|Z/98@&_RKEhYL 3Bq8<G|,4qN<TppDi2(5:hH]ZW!LnYeSùtBx1qS&H Lд3bc7tF'^?=G})Ӌ,PYDt 5ԉ^(V"dƢi1a5F7H*ZO7䆾ft"ͯfUc)D L$U=S}hDgc$^Wn_h.zM,;Oȱqؓnl=E F*`))HrTRu@CbK|aUece]) '6u9{,9Nz/xsP͗Zrsp^h:|lpo6Ѻ. ))k1t99bHNpl{0{D 5hZV; evӒa=EAts֘ [;XܙXϻ=COh|~`D\nrs@܁wTg^A(2#[=VR~c!'pzv+4H2Uh8vJc'~IƯͪz2  3EQ`BSJ#$TEP\]0տ 4M&ݽFB_ԥ*#1M#dީձa*hb=ɗBM=P>CRzAz'\pRͭ4͈q Q ee08! y/(r2p!b@Tvt_yQ W|)cOhADSr(O>%=ځM1@ڗ)\1.O>w!2PwcnSYٷ!Rz<{A<^X:pj`o.=ӸF^Lhmp(ǁױp~w{f$1r< Ԉ~Dġkmqr(jetpD֧fLx] oM@W`Z8PD2o~td"Nn0F'8].w'GxSru8i< H%K! y凚^zñGV=HMy{))՗X21}Gdw?o"KղSKk*WL=TZ}Ns]y\pVZ(UDnIRӤɗi6R^ G4E?RqfMxx{s. '< 0~B6 r84;1`QP;&=^xG l!/ f °r:Ebܳ*ҍ > fTD@Yftr:лC =33H#|p7Nc,n UH {NQ! 3Iu8Xd72Hd2YxIh>|?SgZD6WL8䔜e6 rGOء (4Zqa3s3,/=IF~|Q2-r#˅fW.Ksu8sY? $*e)mxj`gkv~A}&^x0J]/:: 1>;ծ}ֵn9/,lSK$!_c& :RݫܿFW@43(FChp3{a暴Kzf4(\8stv HYn"Ҟ ?(S]{4@q{ IBlkԼ@:DA܇m%؅ei%vbRtI؋ba $pR@M挛j-Ӆ\dPi*$5XG:%<-5 <q4 0B)QAy^qF@hKcEnNl%KjZUJ5h:k+W ^QAөnmQMQC B5{*.BՑ\فP *H:_j%%$KR5ٟVʳiZU/H:-~IKPHZJOB9Bzni$}F8H*bRpNr]Ih0r[wYF}Ny^UOu5WȱC`miIsf^~ Q0   b5DʉOX{ u0~  VgDHM4,^cˤ'[H+xX!5UgÙZZ}̀1NԐzVg625bKDPXE{ |f c 08ubaf~̀3xHg):&vmnחl :6iAr zqub@ĥVmbX'ab9/,RrZIM !pW01`͜W]"Lˠ:1 ev:r^L@ӁubI-k^U kYaP#Vf=`P 2$lB+i"3،YiOk*=RήFP@ KHU?UjVڵ*c`:? pmmLJS2+iP &6?I<@ZUu,iNZt\`Rr4?uc8`OJ;zH0l(N3kѷ17QUI@1w2:1!Gq;s^ELsUO R^z:0/Jb%=fZZ(ѥ> 7 ~S-iZ[OuTRA uǝ Ӯ.Ckkf::qwҵ Ѿn֠w1NT+wkv:jp/Bkz{q06MY| z s!B"ᑯ:j00n|s }[3U.>ڹ &ޒKg#>V̴Y{R[3{bH<ϰ1x]ʲkjuJfXgڠ.5Kk6 L6֞'kRVZkOj ֮U kIUA@P4i]*HaKM[kOT֮5skI=mQڥʮgZ,'5Y).5,/5NKEVRibi)My>MKoRjkR]kP4{ xyDci<K'`ygG)؉S4RVtϠs]n Kä>P?uJ|{qs9G^JSw?BR@L?g#pF^ ϫY[硍BLFѹJ:A竄='g(p|dl[UbClH^R. +Rq {>_#+=//UXgǖSJX聓LQC_@%Ç-L=s5oU=j D}>prA{_k+5,w Wɋ,_vUZX׷ѽo#FVz`]NVnaySY숿o/# 4>,7eWsl]I:BL|-J+ܠWTd8OHTޝgT^[Fm81 m uẽ&Q Hq\1¨5GD&iEmT35\E6k1ͅaUAxiŽ|ٔ 2 i.pK j ʥ4Y3Pj)(OKBr) < -ʤY7c.=l#Ps3'aH+m`o(IXb(_q^#9ǩcV=ZˬJ\|R|-1Sc@Y&Yf02[#("jWJYa]5[31J3'֬*ojymмNuj? 9jc]l :j82`CMi2ub]pFVe#W8Y& `@z53fcR`G:AQ‚eֻ s'j aӸ5w/FRuERjt$KZj&xϲ/),A褪!RVHaF'5@qo4 4Q XRXI Kky0T\ƵfN'UQp>4j>f!a GB5#BIi.LU0TT NʴJ'RѐtP4vlxɆ3HhB~!hӍ`v|n=Ywxb;pz U9FRB`MŽվaH}áaGNCz(8ښY/f"n ׈(X'_ &;!e{:^!j}ϚO">05g/=oJp #Ibh)bl5[a% -5 Z K:G00tM˾RtRSabho44Q RRXI Cky T@Zk0tRUՈ!% jf1a⻆}BRh ڶ:@tWRf:[h:)TcP - [IvbP**BgGF&e1zƷ%Xh%ڦ&닌濰x;V,OQsmX'ajK'ƩjcKO'Nt Ѡ&zf>-z eVQa} GB5?[[#a*Cq.{/I]_}?5^Z@EhcCCಝZY-hhcb3MH>QRgKQDЋq4I7Uy:CpĔ܊gYAj(EQ?'>m CC:xc+B^@zq5qu*p_\EQ€޳C+fw7)={1HZʲ:[MM)+!h)- h),!褪BARK|MHgr `U07ERֆ'), Q\kPG5ҙCmJa3F-JyPc•1:obCbdaEzs4k1 0Q!;&[N /̀ ;uӲ eK-岄O\5};[oZY"q9D$/u\}Du< fr䰯9zqv8>ț JS}=(YR$)ب p.zqEm7c96DJY% \EA>/4Z(У. Y@N]soUWdAC@鱺х!VF(\'3ASQmyXGr_"(9D,`IQuVjH]7#׈%s|aXM2":V޺ 8̏%V JC@DDiŁɡjYRU'|b,%^~r!X\MI< %TQZWKFqZl-=:"ZMYYG>_~U^i%mUi'vxG(G]/A:I1 huhME:่esW i_~U]ͬ"-jߑB y3^,EIjOB0 [0lLU!4_SmRSF R+@]G"A\m4qk*U]Ԛ 7TFJ!ŋ٪t,ҕRW0,`?ovx.ۊBb\\$٤բ~bBc^:iMH">NUQRD U9k]E\ȽּDUD!9J6>e\J4Ł4v)*^J%J6|Zub)OS nl@ܮf\%؝T߾.ezb~pFn.] syrZⅬV8ҌrR1,+})tʼM"DL`cxs> endobj 26 0 obj << /ProcSet [ /PDF /Text ] /ColorSpace << /Cs1 7 0 R /Cs2 16 0 R >> /Font << /TT6.1 18 0 R /TT3.1 11 0 R /TT7.1 28 0 R /TT2.0 9 0 R /TT1.0 8 0 R /TT5.1 15 0 R /TT4.1 13 0 R >> >> endobj 3 0 obj << /Type /Pages /MediaBox [0 0 612 792] /Count 2 /Kids [ 2 0 R 23 0 R ] >> endobj 29 0 obj << /Type /Catalog /Pages 3 0 R >> endobj 8 0 obj << /Type /Font /Subtype /TrueType /BaseFont /WXSZGT+LucidaGrande-Bold /FontDescriptor 30 0 R /Encoding /MacRomanEncoding /FirstChar 32 /LastChar 169 /Widths [ 330 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 840 ] >> endobj 30 0 obj << /Type /FontDescriptor /FontName /WXSZGT+LucidaGrande-Bold /Flags 32 /FontBBox [-1086 -440 1734 1169] /ItalicAngle 0 /Ascent 967 /Descent -211 /CapHeight 723 /StemV 150 /XHeight 536 /StemH 100 /AvgWidth -523 /MaxWidth 1750 /FontFile2 31 0 R >> endobj 31 0 obj << /Length 32 0 R /Length1 8432 /Filter /FlateDecode >> stream xZ}tUGw;ITHWS &=$i3j t T,LQs<wfvgXA3 0?YzeY!W] ,猳m{{wUYrMY(J"5mYKV H ޿ZN3*UVܳ,UY&ݳ!]1"!ҚhHV?<l^ڱPye-#yy˲طcj~7pيm< C*24 5U:"C2 L3yc|k,Ҵ~ n/uq\Åus!sw(5p.ݧ$@Q^D 98Y&@TTQ/$ XpMzO.x(uIv.@l(]'_ޱ|re._2\G=KP,Z=Wh_xHm6Mm;focmV孭uyBK~=z F32G)ӳja*U[7HJExXI cv5aP7+*}]:aA`H'?q,p\^ŸIɘ^> ƎwI t0{S;u/:>ތl?+@HaOî@=:;:Eg{rwuݑ]Vқf5G=xǝ`73 ox",;ǿqg۠z=݂֝gbGvuz)nv ay|QIK _XPzK>< mAqH'F;}$8"f&x\v!EɅ^)_#tqTЃx?=i>(r1(+5*~u%4=:+:ޭuO `C/LSWU\U5Ǫ6d'72p#a/kgi"Vzt| [,TQqc+E]qW!.YvD;hq__~~E${;~ ۳9[,ϊh&IWŚ@Wyʞ`OC)؝*O)_ N~%RQ}OĮNI/vٓ/wI#T3:Z3-uoԛ%7G;e{Aـ^֣n3C,);nH `dfy3,]fh VvVEHN&Ҁ&H9S=ɺ _Hs 39g\PŪY5 AU >6 x&_ |#A;p#Pw)dG$y#}T+4Q:6l\)"oVZΒVnN4FELL0vĤhp[YX-hK! @Q@3d) (0T?Fa5$>!ʳՉJpW"hUxQ [Pvܻˆt7w(#Pʶ0GYaUޚќpj":4D1jk=Ģc"DNlzu,U(WVSVRC-/\7ZG r@#TYꊒR\ՆEQjt sE}(%S_SR:Ƒft7ήWfRDe2*foI!~ɛ:, p.Qch`M?LS@mh K'\7t',~W M翽Szff7лݳ R:c$E^dz}9~{~1 :7]_Dҵy9V03]/ dx1Qߡ\_暈3+c"s~<2!摪to'7-?4:G<,G L,D:O_"*h;mzjr!'ç6f`^=M!R(?+ظ}0vs)G.Zã͗|z:QcIǣpOfF- ԓLހeY72̡8ϤYlnn5~',ZJ QLS5ժRFȋ`J_ ͊rwX[`T%9jA}\4@s̗N+ɡv9:mF魃f&[^zmx_/%uO%1r.y5"728 y1: -+V1lZ4l'EV g[=X4yvpshm^§ 9Mj65+?p~sávoC/\.+b!nbK+ f* IaV #["z.">J{tT<~v8TǣBm{)8F63'db8[?+?hJ`xM_yZahfxXK"]mX[jCd>Cm:+ښT,28@̺/ -y=UǎS2п2[3D`ʥ+tsu üvͥK#{-.]zM.]6d.˸K;ҿGЮCvUvߑ\&?0d?~ky| x#cFJ}wB2BeTʂ ^F|L fK:hh`lJ+5N+]_,ʟ)Prr1 Se~Qh0tA mkM1&Q'q/&{"y$XjlG?dDK6GA(JȴʒVTv @^N˝X%m,FRrp]B8>0Ua5;m+5W45Ϭi𦏕O7MTݤkjR.X3yJo~򤉣"#7o)L&> endobj 33 0 obj << /Type /FontDescriptor /FontName /UDTVVL+LucidaGrande /Flags 32 /FontBBox [-1067 -737 1641 1162] /ItalicAngle 0 /Ascent 967 /Descent -211 /CapHeight 723 /StemV 103 /XHeight 530 /StemH 77 /AvgWidth -490 /MaxWidth 1640 /FontFile2 34 0 R >> endobj 34 0 obj << /Length 35 0 R /Length1 9656 /Filter /FlateDecode >> stream xZ xSǕ>3Wm|0`Cb`  Hd0$G!18Iml_w)2im{:_-IȣI6A$XϽWn|9g̙3gFwܺcg#٨$[мTԯЬǛUwz> >"[[w>]V#68iKzy||34nOA]ڰQo_NnjԝF{"$n#@i9i:%2aL[塄OkB:u๧~}O=6obTӵBiO a.XmIT$S+ Z ZTyq:k܏~ q':0C% 0ngnUY9AڡKawxǹvjϱ} c^z}½%{_x]!ҌmnٚY-vqٸنMr<5h<(56o/oS =2i!Fv>-ͣ:TϴW$(t[+*3Ĥ ?(g~&׻g̮MZP U)ӌgFK+Pmva^[(F)/"Z-jln\*,`)!ͧX4>#>PNr+zEv\R2*e1ߢ0o=.-?gowgʱ`0څqM <' :>Lǔ{{QeԌM؍&%pCS$VʫLgXf44sݟy݂o>,)#ayK\%tPT2IeR0B*5DA˒^T =2 -⽣ZG,$ u0UjH׫q7 _äJ gyfQaS|MLfLoקwhyGRJ#LM<&>*JWo:" h"S:J ˭@D=sY,7r9nRNC3dr u)R[Iα'rVTbѢu_{US*{Ek { ѥ%+2y􈜽h^dl@"^mHJgߋg ; xP!lEM,wwْxEb<~Xx`י@&~fR>}E`~D5="[M;isTSSͧN:{ OI'ĎX tK+=!=8_ű#\9>J9xIS!ݶp6ZǙ:Nkmhϕ5wj;0`j(wݝ lnok2CZ.Փ]"_" -\!-CHcdSrM'" D~  +$YOƻ!BO7H}A$; fN^qrς^sq%AIA\Y@s 3NnITo ̇ 'Ϟfϰ^Cv'{~`Os>!h ʶ_[װ,7GVV 4t$ݐ lFjA@ 5b6l"|2l S|pX6'FnfIȱE>Be0\λΚvntTi'9(b1gs+˅eMȴZFY% O[I dH.Hc s,BWJJ&kK+mYV;~_򓿮YT[_f3*LMH2TUn u1vo\w$իdU)`BVk#]T֡T#PgUpXe +ch- p0ժ_rejV uMT` *08 S Pb[j &GV3/!- &Eh*Չ-lH6BLWJ[wa0ZUC2]@iHSlچNѫEa!OXQ<Շ Iu04 ` OnI;tدtS:ܥ:pH:tpcfx*Vr0G:tԡJjjt0Oo[KWeUe*?CjT*(Ϩ(=7͡wwh'>߄ЉZnBԂo?u)ڀtPuߙOSt;%vbXXE5TCB |Hh~YƻO~YS3XI}7t;Xً׏U{>h?JЍw.bMEr I߇=O_VS0+wxyb~oF=~i-J:u| {8 d 8FJ {M  Ure%T{j*ŸoaCT,9_ PT^ʋ"-l|4Bɷro:.UţF`G]asvYkT"[*E K:EN4mT?,[eŘ!D#  F7>{+زjFIdi+3WݘY 'S0,]:in (ܗ% ڬDPod!%"B7VbژZMӍ7]}x_QQ_*5.MںU`PI]eb`C] ՠp~1"„E|5W{<Uݶ mE5;,'pLmySXЇѠ,JXвiѸ -NH h*1jċb*cGզxDGD?@U`/!/]U1 ^xxXz{£0S5zm 0b&lj RI 6O14HQޟB7vY(X#-yu躁<ʸ,mb%IBWT&E +F{M<yH!l[瓝XkFdm3Gw4:#Ml6Y<1/K9߂W9\i˝]<\4A{$QPWG+ZG1O>SjyaɆ(ywCr*,Wl+¡ 7]PG9wT6 #}ظbpӇ}BDM<0ŗj@09OuH1mT+AI*Gj ;<_&#p' 7ևRe;<{SXqmM*$X=A5c{;v ŲTQYQ /J+"C&Y"4%!xUUaP1 =7 =04l645V%4 ⤚XaV@[î3 Y ؔgؼ 6@~XL\!c1O 0xs%#`0$`kC_O{>>p ~OR=fn\ kh?;xn*\" qd]t4`W9*8aޕv(n{g6&a?pJNC:(q%269Mͦ6)݌ۑў.)ބ\sgLl׊9nLog\eJs.kD4W#o~˞> endobj 37 0 obj << /Length 38 0 R /Filter /FlateDecode >> stream x]n0D .E`Z")xdJԲ + }\i10" Isŭͧvc7un;aֹv))[Qt܍UU;%e>׹K0>{_]z}n/1:ƇrO?.Sr8b}Ԟt6zz4vn}iʛ~=?ۜwE\~识I+}X׬#M,  ޗ ^s 0ch-J}a-3*OH`lH61Zт4WL?DiRygO645u~aR/_XLUaVIjM 6 DU`&Z1`r'_BI lX+gJ`t(V`F/N"Ւ&X"fU$> endobj 39 0 obj << /Length 40 0 R /Length1 26048 /Filter /FlateDecode >> stream xy`U7~dߗfk$MM4MwiR-[""u`YQte;qQFx:6w6O'ΤӉSo҉7Ӊ7҉_#īttbG:1NlI'6w҉t/M'z҉tbb:љNLH'҉t"N2RaL_+1R ϊbxF ?OC1@ b'1|G '1|U _×%1<)/bxB .?ãbxX bO 1.pb8$m A1I oÍb@ aaXoZ(i" 00 pQ~ Вy'An^@p:,@^5^5^5^5^5LTCZWuZWuZWu. OsyG%#=N>zj$'^;5vvLI%b'88+ ɩg>ˉ(qc7B\/Zpx%GMN\nlbۇI.0 S2./c4SV[[EpMV79xu#f+\zŦk3^aɺjK,Z\dhuc{!8=CDmvpw!rۈ8VQ~p r?g!ڱG2<ǑMܭF8Don S Lz*1yDpVqj]z~%|PVĭ&!ԣm"Rxvn07!f1of9uYuxM*^j1k&f rswB5>[cT%dv@#]\WTkM EJh1֋-;2" #N6lYrorLpE$fČw-.Y]G& K(}*VbHߠ1ʏײIa?dg־ x%I@xɝǰx%MO,ҚbPmTi*!ȼ_TT;*8U?yՖY'VyG|}*̮=zX8 'kO<) IiɍP/vPʆFz? |+޶]>, 9, |4 8W>+>X7{v'ɸ&"v ¿œ}[-hfj#^qMtӐ7w l܎(Ѷ!Z5D}CYmWY,SEh-ʐEVfRj9u 8=\+WO)ש#k#FxQ\%R *"w)Ƒ(!^&j^G)YuC!"TM5BIvybiM ms{^?'k<'{\ܩsRrjHBȹ?"G"_$?qEr"Έ=b"2" иt4x8ywuqe׬ÔCiyKZgV=<&΁}@v<97>/gHg|ji|gbk& I&ZEŭ@TkIO=Hg}s邋qET,%kYgk ^o{ ,'U IO8O^; ȈQF]qevg<54w_BkV\ix*Avu% sc YM i׀s;, 3>|>=i)oo`*`/.0` `3,@? t&:U2@P((rTXkؗ/bgb>} ~tOwb&6{5JKbN~Yhpؾ؎ضpPmئͱX{9]I3ynC "!?M!cgi<}hzep)b{iscm9a afE,LFW)fΚ?mbOQTy$d}0n'{td.Y֫>?2@B#NB>&NϣCѽx˵N>C($w w<'rl,K'[H; >hՉ'/k"yB̗/$tD8yZRJ'_٫dddd_R*C$ՙXr5-Iΐ{w@7 MFEys̞;{ ۢ-MBcC}d\]mMuUeE;)'&ct}%Vqi N̝DW6罆xxx0?/u{ o;/xfKs^w bqCQ!tIW/ccՠԸvs>[r* 8tӉ+9k-88L4ʼn5;[{"/k@J485ǩet+cjeR`46pgu]ݽưjUe J5+YV*&8Mka(@ u OxCb < ."x,و;㲖<9Ҹ0?Nv@ 41Ol8?:Lxo뒞xVgW?0#wT \K\ȳ=Qpí.{iK]í .kf ^ ȍ18eoK{ >xIC026exR5u8meQAp?{[xZ <{|;V61a1d2 P%.i7 }TH@0cJ .nz\ìS70_@݉(NZz؝gPLh3<Ðjy:&`  z (b#vH {p9l^ 32,)ڴ!T랗)ӣ$8r AjPy\)tZhZǖ\!lnu&`4y _e ^mL@Q:uluAAr Wk #v5UBէ2 DgI^Ί%}~˶%uP*Zޔt`Sm xYv0j]^h "`6pĺ4{۰'u8x1/ΝU_Qݑ5-S t˴x(2w8u]0RY!kbI!bD, ˒PFBx}2C2IL˰`%Po=zo[2<ǐMIg{H4LWy5՞fV2V.4!c 8icaull tyg`>W\Ꝁv x||p|6ƦxVXW\M:JLm3XC =bx_w){S؋}&Oȼqw3;&P,q"r FЃV \,fؿ %K|XT*X{LZU\YRty&/6݆#]:XgMcL=F=TĨ*9Zo| (K>^Vx!Y*g3׈mϱ LHҟbO2$8g  > SqeV,Vh$ 1պJ\XSFcSE4JlG$ ArA>7 ɼrZ@^l ?*ɳY>X% 1'efT,Lpq\-&)͑b(>v %2_ErOj)YVN?d*G8Bd9}Rx3S7B>GF r H*$[Vbѳ˅8d{27O8RH H6!~R MHh"LrH& 8H>q"6bJ[ cG)#Q&#ñ %KJH%"$B, f 1R :|A\wo_tH$ΓZ%(m b -FKƤ٬]F7UW}Ȑg8lufyƌw--ӭXOX?Ily}/z{}5?ZdzřO:m·YӲNeO cg!1 )x-bRVKK*R( f^IHy|{h >È1l)7dVV4p |eϓʪpyg@F[,6F\Zo9M |r rgK†,}~Mw2~g[Twõ[sJmw KH|X #d_@޷S#hx9njic- Icd gĞN!L=Tf34VrJw> LnI>RΓ'py>dzHPss[O߲m;?Xɔ卾r;o4[ 6Z:s'_Yu[~4mln[Ͽ@;+1.R0dt ii7c E`.܅z(y.Ǧu;9GvHrLƩyg2n f= ?iqAmH#lo'[6dx26NC |43\NuJS(V^o5On*֝QR ,QTe)x"^ϨybMJ.TWa)ѓs0b8$W#K\`-,- }^5  ўbD5;>^k/#!)@NK Bf煶Q%ili6SRIl!$VO ]HAEQ\&+بH||6""X2T/P\kjrGӎ٣cd r;YXZsY5/?,Z&pkoi-3_ߛ P8t Zisaa**regJdt@𴍩F[b0B4c0Kd/ S禠JO[e?ؠk$MvGo| ?.m*ݴV֞w_'N qϷ|aZEOj gЖWf*x / E3w _6,760f|/nB\] MZXAFZ#sX\b24}*E &Z:)a;%X+M\P_|3+5ueO}QwoA8 X=sWsS?jor3tp'9 O,P*oh<8=[}mgͻeY~ /K('jXͰ.],u~$DKv %f3Qľol3z}Y+K.Z ,8G4Ɠl0(2 zġ2s RH6H[v AZʈ.~ƊuҲWMyB[Φ"OiսC7\׹?RI_.oBX}ҝ +<͐'.2](ћX>j\AҶʆ$}]vϣڦ%ko-:o-oB}EU/qK6;R ?' k)+xi6nAjui5h=Z%hLE.>_4k,kFI/L-* <1m~!cIԍΑY8-| Z?8qrV_[5Ҏf5<5μzq`beIYI"^8KτRHR̲& Q.6rq|u)wP]fkN[m%]:ppapgpS8+o nFD&mAi}X{=Ɉ ۶go/uz/ (73Y&''oYcz0eE=ѫnTjǗ>IX-uͪoӗٱAL,}'J*,lep\'\Zx'{r>6l&νpۜ8jq~YkalʘL)v)$g rY͙%-B(,I\U쳨66VnJBs]sjg>X4InNX&eK2W?=.d &.%7&@rM#gn_*xŮUT_[UUA}7Z\Ca C |.|.lw ~&ob>CR!nn2֜NY]jXH0!'U@GE"=fYa1Bmp$$P1REtJ <afe&y헋.zD2,.jdT!S67ee"2d+s{+s_>>-21%zv_ݲM=UE] %5bp{ou3ܫ8l +'̽^;*yByg4Иx}/Իg|ӫG*af$ySudvs/(,85Z\Tt5u`R͐! k.9̙%䚓(0ЙϔE.zOXYr%H& Wh*%lZBO@QHe|./#X42+']ݝ<jR-:G5@AkOX'< &qHFs WTϿsz~"q!N2A /6&9Il˜)lQ@7[w۴{/pb4_؛SMQBRIȝljicf GAQiZBQ- ޑ%ΠE;F@uF#_JR3bAjD),e@e°!&"ei'?exKGA_!,MG4Tx蓥K?eL:lafM(OQ++@ā H }B;-͍Pj1C]ް$KsVZHo/I eyp,KY@XniUehWfeޡsSeAM"k[vkeePq)&ڑHyH-^e,JC"^0@ڇpQ#Xd%vqKs4_l͐\b q eMM/leEw%b;Q\J]dR{;vs-*$q=S J3W:5b+.lɴ^obm0)ϓD꽆&So*ybo|J_L_2&NH{yMߪhUmX6g152Ae KpKJF-fAs0ᩥPA`@_lҗ13P5P4e a+ F:14HzV\DHu4kiŗ2Ny$)W$T [DEKԫufͩTdQdiD]Qm̝8,5# IqTLZ>-JY۰)t.X|.jPIT. B>m<[G#ޔΎ/_Ėnh;.iU[ ;rشb{&փDD iP(bN?N+a+KY2ùa>FQ>Pj=@M'#Lfz%z0,WkI{~qe`瓻JCröJi %_iV>i|ñ>a}~Zt LrIڙRVd'$>?>zXz3p  s+|9"3SgHS4Vk sw` VǐB ݄Bam&|7+5-M%Vm5<̒Ooj$ĒMY5Za5mm-'޷MΕ+Վ%qSL;JWg ׊|93gl?#J3_ILȱ:y-̓;멆( .WH5l5p|hBQ触%yԙh㇗Xdz3Ex ^Z\m ovX57+2HQU,V]*TlfsSSaf!` f6fNo1d lʸdmbN>:Ϋp<p5K3*R w k)Nv2;E߰UŠ\Hs˴EΈL̛d0;Zɍh7=pv. ]rcyEW6V-H9UeV4}usӑ֬ΦHEx‚G=PYk_ҽAКVW3ĝ/n4vaʩ2erAE`\>Je\½ZFc 4kGelETuM7bK :\ v 3زp!0@'%,Ilq|` BpИ qG-ĤW/0$Z{VTgdd9j4qΡΆO, z7ܧԫOw7weK3Ngm`u(N\g*snGܧq 9^vk_kNzaiK௚Ď2 ?TO gI?#/ uߴgdV.ktjz%;%y*8`8 <,d7w28%Rjym)i֨gʠJ.SLWWI؊ ޑhiUZ†K, "'q"=)nc IaZ`"K}:\pZVO;_Ugr#WT4/_3ͨkO|F[ ZL#'Lz^<}d9 2?æ5yL?ԕ-].͐m[lN`]][Zn||ϸ8h /2+DnC&eC* t(1*5J{A6%3oΔ0-w32׬gL+`'3;V}yU}O9:-i`؃{k \ޣedmjwZF(.(=ǝ}ȆM\=uN@{h1y'oCGW$~MsiG4=̏`| #ʄ1YLP> )ժ>:]SX'q#WHŋjavF4˜Ca0why,kbVJy-jCjkhI6pcczc* +dRVEϴ>$dKR`n( GI)\L"GRiy~l>` .52\Yq]T洮Nn%q.vӎrφ"tܱ6\Sm/[M6g-%g~o&$C?ln_:'ع'jcnw^BT|FGI=eLj3Ak#X͒<$h>_Q~'S(Ӵ.wH~݂ ~B<.Le[cǼ◎RԒç\H)ٮvŠk_P{+k/YxPPV]%{Yo(U+{\%swlgOwxB&3Dȼw;zJ.UeIqG!\r*2hs US;\nD*wʃ͗ņ:yW iasrED%,9ĻtXqh硜&G:ԕu5gk2*A%_s c ھT:vn_ƍ4o|dO(H3 ٢gjkzXg?$|#|vQ#xF{p6[fs3 ԽM(v =~3=Q'nO%*N:JU/orQ1r ޿!gЖFtqf~?[Wx`˶6MAqjt'b8|OaP#ɌJ%|H J _.Gq`mnmibgWMK.)3r2\'~+ƃwkpb,4˧dv.nt - >FH£oԀ`UR^m-r5柿T1Ȣ侱st^`ϕi42]@ǵG%8T3h՚ȋ @"5}(d~ ص7"ixS<+yp%xn|U뙝3^r[32~B(BI1;n$*B F8Rz0|`MWkrɶmm*t-.iNڊ?XU4|ѩ굫~c啑\Ʀ_X|4հ}k'Ynrr#Lg.ڷ.F~nwiܳɓqT2?T?\ב?r)S5vHvq/vVrҷ~{ߋsV>43=uXσu}k% 5G̑m,v'f endstream endobj 40 0 obj 17411 endobj 28 0 obj << /Type /Font /Subtype /TrueType /BaseFont /AVOMBF+Calibri /FontDescriptor 41 0 R /ToUnicode 42 0 R /FirstChar 33 /LastChar 66 /Widths [ 662 349 525 525 479 525 391 250 226 525 799 498 525 507 252 507 507 507 507 646 579 507 459 423 527 453 229 229 471 335 507 507 507 507 ] >> endobj 42 0 obj << /Length 43 0 R /Filter /FlateDecode >> stream x]M0:nq6` eBi-ɩ$BhF鐦Քq5B}4C3ή߳̋P\ endstream endobj 43 0 obj 452 endobj 41 0 obj << /Type /FontDescriptor /FontName /AVOMBF+Calibri /Flags 4 /FontBBox [-503 -307 1240 964] /ItalicAngle 0 /Ascent 952 /Descent -269 /CapHeight 632 /StemV 0 /XHeight 464 /AvgWidth 521 /MaxWidth 1328 /FontFile2 44 0 R >> endobj 44 0 obj << /Length 45 0 R /Length1 24320 /Filter /FlateDecode >> stream x|w|[jaɲe[eC{Ŗw8;;v;@YJ!PF(@%mib4tmZBs1my{:۶l_N HMD))-ٱU(VlZ!Dr,\sB0\ ,,W߶8K/ûbOPn^%ݴey,]GuMn-ȹ2D@/o55ZoRi򾉛?u$ABþS~G|O!E M?rMDQ9u 8\*.p}\bN.vpm\lb3.s\b5X .s1˸b\,b\,b!\s\㢗y\p1n.pE'\sl.Zhᢙ&.h࢞u\r1.⢒ .ʹ(㢔.(⢐.".\dsE&\s΅4.Rr"d.psE..pr" ;6.\X0saȅ =:.\hPsBɅ .>#.Ň\|?;s7._x?s'..qG.Ż\\q[.~ůoqq_r&\r.~ŏx׹xr*?.^%.^\|r. \|or<\| \F1ƅpcsqԬ8:`DqNQ{4 f3je-ѸHk]j]j5Dm-zF!Fuj6 Y,T Tͨ*j㨒QEڊPy*ZJYZ 5E,gaJo j)Q+˾!QU(U(Q?jVJgcu:SYe^VQ +(Q"eu%e)(ifdbddd`9,EjiXN%˩`#<#I菁"C};>iA=K#;ߙVz~kZ kW[%& S'k=?2zn\yxƠe%{]@۸q[UoWzGo_BSUYf3- [= < ,DF\")n=y<< << < <>_>/Bu݃#n.N(pp;݆nܢ|Ns˞t{k as(|0|ɃN ?gC6n_xOx=]'wϋבⵡÊq۷m..4m "n.&[n9%EQrqH SZ}[M'77X^ \]2ᑊeኡҊ𒓃 ËN. T pd_'`/ v;6`+l6uZ` XVˁQ`X CR` 0, X@ z@70:hf@ 4@P:@P T@P @P@ A 2 t)@2D ppqV` Z@SBF Ic#? x3'G;~p%& π?~ xxx !*e%E | &< ׀ sWgg9`8 << NQ`''ǀGG/ǁ/_wnnc udp ap5p 8{=n`l[-f`k5j`X,F`0 K% X,~`0}@/0 tsNhڀ@+4M@#/w7~Z1D;f$D5d+9ב93:FQ!_#!?cnzb2ԥ' iFOc,SLܟ.1e<\(&LN}UESe4,^m/.70dYLUd5,#9i+W KdidHmd;فB4mNv.!{>?Sه=r. We&dř&אkjדȍhq:9BnB;B[ɭ6r;Qr'| -C#w"~YE&ϐo'yJl,B&`+f9m}(#fGjI3kZX ĭ3?Oj#z\qK^؋[./3sw/b>OjUf~Yόo:q9K!eÄ*,&`l?FNS=S'W䖋q%S,S-s:VWts4Acl/G!jjHoS$To^/a͑7,Hmz._O9R]LJh] G?Wxݩ7_>@[gd] (½}&7όێohk!ۂ_C , m n³[b9$lhlw*:?\.#&.B5r 6/oqWT|u1c49g{HDDܧE[卐~r߀}(46ܾ>z0(v]b,-BR;yDZrfírx:83m<;u/*$^ 4*6|VGky-ޖቩCMC0 |mc<Я!7a'3n]/{o돊8jOGZ9/!!9V4fi9\rB#8X& ddBdqODŅ`V y 8є"DD_-rUη!5:_U4^kNƙOO|B{NLM4!X1(J1az$NGfY"ZԠՀ-rQhmD rp 2_ڿ^kپ*4;S_?0fюzJZ\!F_KHm42E (H/::nY,s -ɯ7"mqEuFi֯jz&<.FzW1ۆYy=b|0iywn%þh$sd`Egcrؘ 0{iLk6c!DAeħ&+;Ɉ*NKBE >&̮F1VTy%ǤT#>ART 0'w;`%P,h@̃bn©$ F["۳xq`%&.DI2DB{ q#jБ8ETCMk|b'}Cѭz6Շ3;]yAe189&kkqqmB.o^TU.\gHNR%qbG%Ut:P&R\ թmvBoXjlZ-y- :[ee~5 i-\*Av'd~S%)U]#X߮UlIg%%͂F* )DbK볜nBRBhZv*zF4f͓{x\"מ0`Kfs2w &Dk(%EWB"Ȃ\E.d)Y,$b.ϜHHIdCW %6l`|XO{NB s:V޿%B_ZKXs( qy>IErylydP\VTD) NMO./V}W'SJ%>e<.>-~T(5Ax[ A ,^qX6j)j8H$%f\{5R)Op) BVR+kyqbPa0DG\ cx7o嵛XM>4+-Y)OƗ.8>:tg[.֩Kǰ1cP Ȍ}Q?!愂E!{Yb.ZmAAܴølq["vB?eܐAʉNVoPN+}fZnuL'ʬ^焐F%+|vAfKyta12UcVI#;߸:^4⹛+Ӌްj^q j*(j}~`UܒĢ5 k}nMs=L_J9Յŵ}{kNVݖd& esjgnF\6K1GB^s!AkK p 5x 5s6@ɄAD22D0uImOdMUf,7xՄ YIJ,%b%%y;d~%MHKS$>.ɧ]%Ky~>d2 ‚%80K\\ͱEAGCL,.*+,ID֭=YY8rp[Av7_Q2|C_࡛F<s7r *߲sSdn;ٗ$}sZz0º^;8$N]Ej)g`)AM<Ȗ.-1!C|`J'3O{Bt3Zœ A5„Z!Ed3rЗņp0=ؓ']i|k#!O]wWҾVCW#J*ƳYBcƒxΟu!=E_VɔQW;:⴩ :sI-2= **١stҽjkR]>[4rӂ̢e->Ry\ ^DGYAP ]/5͍!ͽ j 5Z>˖&AXh$O6Iv/eJv{j9Z9Ԍ9hE_Mbz9%X}L W_jj * P$ёQ&T-Hl z$Jj}͡o||d͝C';g-MTڱk~#ѡ6%،vAسM[o?t4sy9ZzFۤ䕐ÖJWfkZBgkqԇԓMSwCfV bڑNDt1BnuqP LJW{-(֟~sZhmcBzfv'~gk[e.q_)_ cS :|t#@ #"rI']8S[#7%  !Fj_R9#n[|}7+Y{bcr,"'d՝Yb^|5?~K{fX^떶< IvR|yNL5P])\ V'$t j5x-s`|䒪:QoL.+cSIǕI}DȶSߓVԉ'C)zhtFOt$$`/=۞ޒ);Z$l"kXG5ee4$ؼ3h\ymLpyjEڶpogjFQ(!4'e H3h&6kōG\;У3t蓜El9lrEXr<|Z܌Y .ʥI̤)'ލWk$Xdq)[kVLijBg `ϩ RL@ްZ(Ϊb-/Ԣ$t.;~c'1Xr:@@pgtrco_kr>*7%6 y?,tQᢟ -3ztC#XޙDh'NNW?;be 茕{YIx6vcѠ k[r+r;;7ֲ3 ~M~lKB6+:DFx-u(_aiʫڬO9yۦDžʖL;oih*t/nW15Vzpwb~}faSC{`92s&9gṘ0#>֚=]0B=?s1wA;s+ԩt mO`lcd+Meڈww/ C@CԷU웰ݷ>J2mBUAhZȖ,Q0Xv,`y rJ):A7cL336^ugͤk)aBvllq%v03o{\ͪ_ٲ*>\vMwTwݚko7-vC*Kj?4\YlsQu&g!R1`ypҰ} )7 Ү4U,@] q:ѥnb׌ld#/lsGpY]=d>(bڴvd9!})-ds SI tPU% Ӗ:ݤRp^DoΦ8%ePl]SŻTE^ڧHXSRJA>TYӼIV'-V!$T*l:ieR@8'a+6I7E#%.EW!篘LU!E҂FLoi_8;_j.7PJiPYiMv5o],ed@tg1ˇE*}3k|B]yć=QH;1'eev Ltf󉵣m\8Z8rlŲ[skƎ'W\P|]vZʦ<^swxa^]]VH4,.k_߳xI_^*.n*HXh|0~V-gxZK/ϫ&|f~LB~;rysLm8UOxOlzN`yMJW 裰B:E|]N$r A`~ f ) :V8BtA?~ff|>tL!>rW EGL`MWAx,gJY.8v?Ƈ<=; sozfm;X޸d˵p_O`ޯ'y䷡!=YHO|n!=QHO|,yecđ@ Դ$+^v}jJqB.=<8! ӣq<&ӾyY˪KEupG/⼲/}I t6T]Z*;,*Q=XmЫTZF0}dǜklaW1iMt=$mV;u cJe18aǷhtkߋ>]ݾ1LYB EQPHp'N9OTl]yF`p1GO[+*Jt^ vmBlm37yvG668Z z¶u0FTj{Zv0%Օҹ%,3RX[k׫lzz/UZq> | ݒvxNMyBX^)<#3Q'h<ҊN'iϿT$; z9m!?18jaTXNtj9%)yJs|zbZS;vΤjrUm/ZIev+8A5c;X/< Q@ȳ!{VB$dNh L'MN`T6$)* WJBSJGM&/F{ t@Q{! W e-+jzB L~U݋^W7')kP~YS^Lx8*JQW*ʤq={ω}yGsңC(ZЖCSi Cu*6t${#͓L4)6sǁ=+n-Y]h9xj]A~hi&?]f3"&Bww5׭^e endstream endobj 45 0 obj 13442 endobj 15 0 obj << /Type /Font /Subtype /TrueType /BaseFont /ZXSEQX+Cambria-Bold /FontDescriptor 46 0 R /ToUnicode 47 0 R /FirstChar 33 /LastChar 79 /Widths [ 592 592 639 535 591 308 531 220 592 232 513 365 314 459 469 326 569 461 597 604 597 890 597 591 531 597 337 614 350 573 652 662 705 846 651 551 578 679 592 592 592 520 408 408 232 695 525 ] >> endobj 47 0 obj << /Length 48 0 R /Filter /FlateDecode >> stream x]͎0Fy /&60HQjXG};4R Q ޾C,>kC0ݧnm:<Ҕ/fw~(K};ɞf,4o9vCwvUU8Nenc:=:<=|ۓuSf狺v)wl4K푮?ncvLDuel<51uRyH LEU>#Qz>ow᠎RMr_W@>I+j+AW$ޯ0+z)\l+$le H-k1vs[xHj V B&k!6 Q-(> 1R@MS(B@Na5 S@ x7  *f*ɀ!r#L0rv*Jgƈ)d hQUJ0"ҏ)2E. ""m297*r h"ʌ&[ v4qvӵ0Gm`22g endstream endobj 48 0 obj 520 endobj 46 0 obj << /Type /FontDescriptor /FontName /ZXSEQX+Cambria-Bold /Flags 4 /FontBBox [-1110 -299 1372 1047] /ItalicAngle 0 /Ascent 950 /Descent -222 /CapHeight 667 /StemV 0 /XHeight 484 /AvgWidth 600 /MaxWidth 1380 /FontFile2 49 0 R >> endobj 49 0 obj << /Length 50 0 R /Length1 18852 /Filter /FlateDecode >> stream x{ `Tչ9ξY$d&w$3I2 YrI&!!,IH CX&0@@dA$U(.OE^\bK}i]Zk7h ߹3E9ܳ|9qHv",-?_cwF.1<{g*=;?~G9i2ށ7nI[l W߸4]x]4S{9֮"QrBrK>@ QA3b\鏩"}H S8RG/}PuQJ!~RnEL(9҆Obg6&ل%0flBMh U6&ل,f 1~_߉y1~[%Ư91~I&?b|F$Ƨ91>!OA1> b_o=b<$ƻV1%;xv11nVGD5@a5v@QOB!Fy؋" ̂; ('!6߾c:^{mXny|D7hnXm߰z܍̬cJhپt`Ϻܜ r<[!_HO8il}Hk^/ 3. 5 ~2\lb -1cÆHJ +k'd]4]Hv 5 ?I}kܰa D5uH$wP;!8 X!X ! !h! (!( H!Hɋ/'K.y>vWW/%ilLG'O&&$%$&w%w&w$ۓɎd[5y;l#F = b$e|w cMTW:~ʶ#@AoA Õ~#r )ގF?% M :Ps, <I~\jCIm/hz=Dw hujKq P_W;rREy$. ~\N=7ffuZZTeR Ma6ۼRiJ8-hf^ h$^%H2  Uf g͂?nZR(p8t+6zRbo("`'6 Kr읪k9kN&az! Q{=k=qE;{OUM\RDd*5IlXSi9A!g$m&aH@ܠt4KWBPMLaqLA$ӓp~p>xAAMK~ DG{<ȂI@H vĈly= nb.MO,M4 .eʦ޽v{CPBsf 7y;C؊CfFΚW6omIqsna3g4y`w`%`d)+Kťn^#4G=z^o;; Ӿzɢ }{a=L:s񜁲šNI5}hll:Ñ Hl3>$Jt y/zf?a~o גn-Éc;sn> mNEǐv`\`F!2<,&b#[5p*0^ȵ344@hn ^n!T*T M,!be'TC}+fge"ؙɅN<BONh1;;{n#A)9-a{Hlxf#_q{ՄזNw?1qMM"U EiRA< ֠ؐ80ù_&(HzOknCVH0y&| BU}k#y8A˂vr`ғ%k3= #+| |V0Z]tD N)&ق;YBNʳ'I * A{%**Du0>Ps` 7Cn>Dc\Y>l`l@$g6@\Hi ߀eY 3Z@y83a.^Uܖ-~3Ax:Oxy@Mv}TS1~ eܠ`\$I&&/CoOQ NT"fKPҕ %`yLyt/<)6rI%>0`#Kz$P\ L4[(F@2/8'ԇ?0^dq)psy=70Xe0iPo* Ȟ*\OA}mKbC&(e"L-d$dq w P+W#IJdH|ȕDTƘ&2B10Gk`K8$%%~S} %~T@b!"ˢ}jJPC: Oŋo{3`lFPÌ@iЁ62ɓ/n:O.nECɡXAYHWCz_1,Ǚt\#*'FH'^G3>`sbB{ix)Oҋy*rL0mzʱϥPBrZ.6<<PB@-%wvȷn5JZQ0}I^G=%+9j!F݄jyMoH}яN/myI9b AXN=V1tBPZ! 0!% V-|P1*Bz (<`{ ȁQ>E&B^D(ER`kʐ 0T|0 'LOCT ~EK>nѲc .kZ[KQ=:F}/L+a1>g*7=g^͚v}myZb;E6i7w9*Y4Ϲf?2kHYԑG p{%+\s&E\kͲVRf)Zvzw7GԲޡֆ~Vn_-_QE 'ڻ2׭3R误k)N'28̞0,_: A#zz,RуlpRGt!gvc6ZeU*UbJd (jFm Xù FT]}feFOPjƱG.f+PE$8UsXY\%_T| ,*h/Z,olO̽Q.ʛ^K*ǿ=NE>ć,СK!]x[c1 БCJ^}`4:9T<(/E(|DL Yiȿ8CAGq#P2J2Xri;FLk.loI]H,~2qEbd硑Mo. 5 %_jT}]a"i;lZֺ 4|'SK= SѾ+tl+x:E+iCCb`c7|lgb{s9Wl`Tc3+^JK1VF 2*ʍhE2emGyGJ_rƮ~G:)RH}]x\=87',{Z=p̭8Ca?sz'<̱^=8 0_b|i%2͚e@nrDL€3H`507MynʇS5YVWxNK{>pn׃ZĚ^m+j^_xxBo 5Y<ю+5ZÃaXKxEأjD Pic-<'X̣ez ]|^^2e7+:L83Qx; 2g BdJ)MWUC/(=%~OPGn.?z+svqRav=-j 95u5u+-PkC+'#ԍ"VvgpYK˔ 屘 t.. ALOuV.ZHaZkvƔ0`V\j&iֳЦjlTxkA'su]rt9iZZ0M11[u<$25دɕ; l(*:`@\]Gԍ"G5-Qz =Ko֊ԪFiJVߓ*XfujhˣbR bI+Y"/WXDQ.EF"[&q5k5a+q_a"puqXbIa\Y#KV_X(ߤѬKMH!dTS_?,ץ&=7Ѻ>}cw&\0ch6=*V KU^ dl =Bvx0i +aسQXLkt5\.@(pn\D2<3R\q,& z<4r˜!Eq\q]  m U *;D,*ZNZZaxF%7yw_?'z"[2jͳRɺ[V l+^%]%=;t-l>PV<S!#sh] k15ho-\p(фf4Â7K:.ZV~Zgq#CaCEtz[QI(7/TniH-kVDnzw`iġwٰc]{<^asNžc丹H}zZEه_Eݟ *Eu'熢#G(ú2쐎itti>Ř:AkE c^/{ BfV%4h։HE_}qnƟu[ļ>.p iQ0 u)7u孁MfWvԴ**ktV-s;zֵ/>P御k__Ǧ6B^ׂ3=v~3jJ.ԂlE99sflfI K4}ؔczv2LINW'z=09CF]2(Roc8U(؏H~yoV4kRP*E o+*:kHz&ݕ-l1BQn-RBQ/J=]F%{Ee>k Q C_H]Ȭw x^ݡ֪Z-ѳyVѬGbG0,A}Ă@F+f Q*-&֨h(&D *,Xau)0tl/@č_Q# |pD QOHԣ1pi7ߴ0V,H=p8V*-hmXv,,-tg}0wHʱZU#d@Ү0COz;AR@h^ {føީ6Ѧ#AvX^~-m(@k@rYhDH旱*DZUą,aEu`tnjoÍgU+f9cG۬ݷis`Oðn}>YQ9s!76~͌LA+dґLTE6HZBiP$RZ!;7!~ĭ.DDkǢ&*dMZmE1^&Cj,.lU[}o꩑eݞmWO}xN_{UM;q۬/.5wC¯5a)4˴$&V;«tM'0O[rq(77me eH$Ӛ)p m܉|a`e1_ۃGĭJ(29e?l}~ކw:IZ;f)jG>Z^9{Uyn'EoXNe&u1wGx=vGbRTޮHtxc=OH5D)b,hD&ȳ+Ao5^[I<8E X6ʂd9h_ig(Æ^` SF]~tjYnpϜҽTY^bG`O؛%N:psٓKw E%]yQYԟʗ#ut>03i̼댭3yB9&\"$SSY>TR0 ̙39f]fP%A:d` M͑95〜xA$Dm+:`cL$|!g$%$߫4V~>rl;y)EMΏ Non6m۾wpgR;cOC귐<bhp\׍]2c#eR( C9Ü2 {b.\n}Ֆ|?O!_I>2'M3I|jZWJk&yL1ZQ +@e#1YD.*CCZm*/jZ믮b"%/P%+n(,j "ݒ.~EM&y Oy ?>w\5M:=>}$nr)BCU*W_5ux^H9_d?BܢluK´54402ezIH^ډ$Ґ$CC Mߡإ;;VҴ*j#P$2/ sJ8)i@ 6ҽtOmEC0ͼuON rvQ[kruHjQD8]]F3RtnNH 簁>ĵ+30_DdɮiE6L?7i ^g=fl,zH_hdHr8e{2u4HyzM]wQl[Sj9HjǢi;_dޭsR|*%Q,ِh. ^hvlnf$#0J/pYI@ D*qGx&_2mn䳛M=(Q15h\ gS^/_8wmḰ7̼ 7#7q7KŹpxe<:pX *d'6AVZUGȻ4˺DŔ㨞 Z2zn3^UOp$:Ok◶3ŕE/vp͐18Ѻ|U=XRXSմͫ7Ԣ1oNO=kun|d)ɯf732)eS 7M0/׍Fl)i{, 0gD<͇񦦡{iYFnLl5fheWISwv{{agwl*oi{TYUдu\>{i@fCӊx` +w5 '8%Ld~bD;n_5^arneqK<ޣ/eo!-,84m*ИMU mj]8˺s9YmūjQ 7@fUb(F9rȤЩWyT,VdB!b CEe‰^=u3Ʊ)\g%B3J@Va,+0wh3S/׭7x|`qۭj^Ѽ|T[)oH]>`ʋ5,ȮB͖/ilTLOwX%4wpZe\).iq |47CR0Gb 1z9?֪j*-$v,3DRR9 wΉ~ASF FteW..Msh9,^c ح&8u9eÞ%SjnFUuSr߽| rp#E#ݳ@')rC\ݘ+~§d")sPSZR~k|dS3y d8;1"~OdW;;L0YpmtUWr ,)\{X-Q:繏n7rh_^#5c؍q+WzR?{^Y30`juv.xG0'ݩ\LpZQ&^H3/G5sRoNWN NQ-[ Q^!GLQzft_j \]?+6GN[FpV"6AP)%:z?'Ԝ@wحŊ+B&f?.qݣC7k6H$ucv|Ҁyy=H}A```P>U|:(f=ܕ( p(-_kz܀(s4j9!v"b[E-)&ZG.$@ HI"&T|I 0"&F-1FǶ-{GϺ?|gݯ;q#xfޜS͇bJ9w.B˱÷Ya၃c9p>C+TdZNF5"f킮¿O]w(0zFA2x'<l/c +<@r},j6KM_L]+GcIN8<=i5ŒARvGyU>ZRbU!W9bw W2\RRq0谒"ɐoy Ho8P@fq_4E8uƛ?> endobj 52 0 obj << /Length 53 0 R /Filter /FlateDecode >> stream x]͊0F~ -LȢ?4uFqyS>K!)e1ØbtAL/1U1,+UMvY|Hdڶ2Ae7)N|wr> endobj 54 0 obj << /Length 55 0 R /Length1 13692 /Filter /FlateDecode >> stream x{ |Sיh_vY+_ɖ-YMX/ L fKSB N6`d4 Gud fN6%/˯3tH,- \ef̓u=|g9ޱ1FȖV$~̷£dXuB^~ !w+- h!.³~ y-;n*P|Hn~-yܖB!xn+ҩ4B٭mQx݋ 9"fL,F띫kL r<_CW^zY"p>N|!>xD*ځJfT(yjspV[K~Oy#ܶlZsӵ{9v`oۺ p7;](nĻ S矸Mr M.s+ P Hgxdw&.L'|ptSPȫ#I'<~z2ԑt>]YMSV>Թ_AWf[l& r>ipb}ܽG޹C,tPEy*%>zNf*)>[YK&*x`0^ )bX,Z>nX1,j [)9@zJJɐ”E#j(0\RÓP "rBg(QHG!RQZ~s{B뫢* e#)#AT52B%R h z(Q5ށAS/S#5VP2W)w}{?CӔF*i-TR:%ʓdU2,OeI6N:%iLʤ,(cjL< `) 5>Aٿ*=CQR>pBC)ҧB</+`T[&J3B I]A}B̀ڂ.lqCR UB#':B9jL]v!G vzB+fQ tp0:'We'Uk~PPlFP: $6.8~=?͞'pGAARH](EHvbpy]>0;d)Ir6W0'\RІ)DF[Rx.+_,:6A7c+Zdh݄V`NPJG=(bR͗>_Qgrb?$g`oEh H# f-ԇ??o9^LLSbxvÌ^w,t7݀@~?Π>At;OȵJg^Tv[?' ܞFۥ HĂD@T[A20{|m*[sU+3Ctwu-ɦƆxM, WUr+w(۬h`:FR*2 Qvޞ(gAR %Gdtz9diP@>ܟA|cF?]eox$~rQ!0r̛yzi)~ 9vT`bI:5^#Foi=͐ Kb3jєdAd!7$XD?)7(o$PŰ mtn:F7˘/ e'!zI1HϨU\ VV ЌJ %jRl:CiZ(1Ak:Hd(RPL)]IBЬP A51Ec vA^As:΄L:Š٠f I \}+&Wv%N7w?Вh\ٍd)1qĥS" Arh21uhn %AprL17 '$(Np)O4$FsLX(gS YTܿ_Iҽ/% ;RD:&ݞn>gwt_|1e:s$BY `Cn١jECb}׳vN3;9ɱܩىp3֎,+?Ot 2"ⰿi`aU';rF0_ȟ&ut$oz X''IL DύNi$* Z!:F Dh ;6,+8P,N^OOh-dt!Ϣ rfbYN(sYf'1(N3 bav ?izS{;iRR*!)؂bC XIc&(Hg XRP*Ie"vJ #}[#eAUR{t|^ %k3ȑN~1;^KDz3;Ao lO`@sf l6%O;M*΢ZCYAAPf|̉u0,}(]T7OX=d)WQ +DoAuE.|e`fW&ăD_̄.1/?E#y3G6QfxӰg )=ۖ-}EKiRJ IdP=N+!N@Ci"ذ;L~@XSRX<9'M@YyO`sԋ|r!˨JBUCpEPK( 9L%uh߈VHbWK .Ce@KIf?m@/P6Qqc[8D$KZȊpvnS5?)C:,Ⱦ)_ !oT5R5kº7cf==S:ZMfL&ӡP0Z>6wFQRu\cq>5&"F[l":vD8dbQnhu\su7m1CF- ^g+wQ_2sť%ClEnxIsk*J"7hո;+[;KvI>ãlD> Hǐу`ڗzVK|X*zjZmcHa+v$-1 )[p%쭶x"{Uǹ`"o[H% g?[goC ]W[QdMN Tmrʋ8:n|wCoL_44ҷOwKA>jK1Ih _}mHI*e%C-e1E|ݹUy;~>QPQ:S|UY=Uj[-SV*8:䕪)%Gag (ͧ@9%"bQ'3E3t4#W4xkHL!29xM=!s޳zF]]^ZOZõjߏ8=*MF`sz*ܪ%!'֩Ḱ:?6g|vkxiܝmaT]7kIYX斟4Uj/Vf1n%,iLw``{D-,h=/1BE?lY FF#v׫}>첒ktn+166a4araoDEQOhh"1C'~u0 1s` DĘGΉfb'ۇ͘qHEDh ;)Ym$T;Rkl{_iJBe|K]VhEjuc/5^Pw֖sUk֦6ݴt׳]v\xP%50va^S(,3 u`u)^pqXi0m̥Lr/]#?vǒLe[l]K׻zf_1x~"j cYGql K. |6`D'f-NfkQSO@ AKgA -y g(Q@4``+"l,~Dc^7GVmk7ѯ/J};Uo Y潢IQFTM~JM^4ة)Uyau2sNvp"eGpb' U.U44ĩy0>-keBSҟ?Gɭ~7k&Z[>-%K%_mv8|!\KzLb8%*pa^ w8λh5vI$%W L}9Ccd:p I†V%UT^&n-ڕ~#უmDc\gnዛ֎<ܰE 74{;AT;!}r:up,uVdaR_;K\.Ks| Y0cGykN}1z9BT -p/A26eEG\ Qsjq9/gq?%fSۏHC1<̠*pRCi=ɜLԴܶ 0Q8pwO˨^ha #VL;5Vbl? 'RVF}"ya%N&0*d(9Pp.$4]t: we檨2iQ^T<¸ˌ[)uJl]ThRGnլy7Ylp`GZꪉuZ݉RRJdHlpK#pd+wC6rNYC_(Jt]"yp!ȭ{ƘRS)Ӳ՗_әQZ)i.-񆪦9&_78p¥[%7TZe]е"&xW)b>"(# ?_U0%7/W_hBBȅh 2^rk( Z@DmZY셰[ʢJm*4E"M8)J\9̂DTB-3uQ ,WMj[Xi88EOL,eF1WЏ2"˱MlV<:f]47g&2ax-Hg[ɂYk9n4ocL'aw"e/7W}Tf W,=F1ap&3ݛ*)h"LDi$-6եҽ*i`K[b%M||dgi&Z[4RdQM./=ǧ7\qUS\Xmw'[ax7`BIT*f#mφ褋Zy֍6clB^Ƹ'R.YX;Ί gqgь=.FqWm7ُ >rꦎH=@5[/duXKhkx,ͭ]݀>@x P6_֮MnQRfcNQPZIxh!ދ9P01((D\ 0Y{=:K68U2&«ܓùQjH%gpWKWvܙvUq8a!=FBj;x]xC*e?b-dV,F:'! .s*iEu OQ4&<5@P|(x'$>21ђt`WW=e F<) endstream endobj 55 0 obj 8423 endobj 18 0 obj << /Type /Font /Subtype /TrueType /BaseFont /ZJSPCZ+Calibri-Bold /FontDescriptor 56 0 R /ToUnicode 57 0 R /FirstChar 33 /LastChar 51 /Widths [ 606 537 399 347 355 494 418 226 563 503 537 246 561 246 538 474 537 537 474 ] >> endobj 57 0 obj << /Length 58 0 R /Filter /FlateDecode >> stream x]n0WYZmXXJ JZ5nm&UɒrTh!))c1{B(Cq܇bU *Bp5aW!K@nn09#NBgp##NĂVߛ_p5}JxK~7qBu endstream endobj 58 0 obj 354 endobj 56 0 obj << /Type /FontDescriptor /FontName /ZJSPCZ+Calibri-Bold /Flags 4 /FontBBox [-519 -306 1240 971] /ItalicAngle 0 /Ascent 952 /Descent -269 /CapHeight 632 /StemV 0 /XHeight 469 /AvgWidth 536 /MaxWidth 1328 /FontFile2 59 0 R >> endobj 59 0 obj << /Length 60 0 R /Length1 15904 /Filter /FlateDecode >> stream xy|TϹw}̒LLIf/K6$@ a3 MQ 6XZW-uTZKlij*uM{}fB~޷c{^ƵCˉl!<[o5>^@uk]RX?%DZUNul\isز-4|R L"`z05.Xt 7!or |f 2W..B,XV&V_B(h#DI8`[Jg|O,6V|FU @yz=|v'V 5){/_?UTS4SBˆp_SSoTpjைb')G?(CT~$}{?"8NUC#~! āx8%N8%xMh:^Cx2ⷈc ^B"؈"~}-xI@Fx# ""@GC#`7\#G܇؃Y89LE܃~ Lj?B܅؍؅qv6ĭX7#nB܈徏؉=XZ5awWcmW"@|'(\؂ q)b3ňM u!Z "j@8q!b+s+KK}^b"Db!bb>;X -BC":vD1тhF4! YzDQFD*D%b11 QCD1Q(@#$4lC-TBd"2^'l#MmJ‰HE D"aGVwGeŒ0!B! BuJT*r #8E t1C|qoϥҿJ=S'AB|3b!ğ'#1lĉf=wakE[k NZ xͰo!^Ū_A+; %XE G!rªA4)ߓak5xo8XacG !"ƪ`!Ax}0boBEqbg{X=aLO? [Z ee.ரEMv.4M6L݊ ~nFA7b#v"n&}-G–v(-E\'tAw ݀  =&U+1 \.S:Gq 9 gd$ g SnC @;o 7r#Av= ׁld;M!NK`?qdKqlE ٨D\X@\Xq>"0#L#€Ї!(ChB(zjgQA>I; o& ayGA9ra; DM u!D 0!"e "py.,:~~#  y 9{߁ = ȳ y O<$'@" Ayd?>X1f%a3h?\t r2RDуXXFt!!AE |lD""<t#d!(H"Ay2oA%_ W畼y;Ӱ%x-6/ݳ9<}sf^xolV\Ұ)xMA٦Mfc=Ta(9tb!>ashڡCGa1nlz뇸52{Hk_0\g0(,:HR.o r`o0=YZMy Q@p`e_6c W8_ذ**Jqr󚁃8'+\_\gYpI\ .\\g~;={p`Np[{M`㞆`[%NAH*Nݒz*UMY­N9r*_|*AI%HW3qGĽrtVmV_sD919j[W<[lR\" F({r5jvRJ Nnm ]LInߢ5Ck֬Ȣ5Y;(||+&o0 kh!Aj `&R;}"S^^N.'=0 j]U1{'1Zh2͚R?BDAlL89~Ix)m ){F371u -1IeMܳP)::q涱kNwuVVA2D6dl&d+F&_\אkvrA'#7FrBn% w]<Ie9w{}ɏO=x>sСf7!hv>r?  }d? ӱT&Hh$%nO#Iy<#Y9J'/'ZX /_`#%/Wk 6y'Q7_歨ջ`r,G&mބ:%K5TE>LMRnȢw+n,{!"^g|~?ėE]v47}|4VC`||<1c<:g%?>>3Q`>d{ļQ ޜ=AVپ谲̧2,uHC4R$> M^%&ϤS/|B>_As Rگk>'/iWdlJj5#cB)(O\Ѳ*#4US Q=5P#E_N昿sԙ+V~񢞅 ww;;涷i0zXU9bҒ?7'IҜɨj*B.9Jr^Wy\@7Erl냬,E\5K-IKjrUW \: ݮШt"]˼RB JuՅFY!#-\j*)T2릍pDg 񞺾eZ-HTWHQRJuVHk#&ק[&,[0_7<-dP֦vpPP[ а湓7!$?#xaChM_T>#,uqM!&6h!fm&"%mi´,qu^s8c -xN뷇,q@d?OH|W.Yط|X/I'kB:n$/}Љ ]: TAxVvtIEP[J ޥR@!R7ڻ‰#E.ǾBRDY;Bnkيױ W}Bn%:P*}uRzT.whU_BudB LVW.Q vuV==5 PEkn4ɁfTmA#gڄ֦5kPnyU)$Fk/΀&X8Xrs8vA*A?%"m.a-ۺXp6w Pvtt2 wsgW,Ϗ]zWV)=KJO&v avs!Z!q (}הdR\&Wp_db˒Q^]? иlX誀XJ~cui͝չ9T^1e]Wwv9xI.BDI1-S2KBB%;lreBJ/pIFdiC)fǁN:QufBpk %ýlr+h $ ̮Ѕ4Vf*BBuZ)8'kp .0:LlsWdb}1)d~WH}@iYLzA=+eik Tg3qi7̅X`RC h `Q/a -@[ vd-rL! Lcr/Q{8N(`LC65I$ \#ZT.,퀡k 4aIyKqD3 5!*?vCKmQ)y2ZY- LiXY[)!;4嵠b.XGPd=ׁyOgdF>9$0I׵zI=</~P M@5x漄l]NYfr)MdJ?K,M?'\*k=W&4rOAfB +xIQ͈)fxd.dI<3f9sW &?zjK٬VZz^Rddk-b\}jA0Um-v<UfƯ2jVPnkYgYr֨ɴfu.Lט4 |nɐڒy[fYo,oOx~'Wϑ_<%AJ ip EBWJysw]PZ%-5VO0E K [ ˼zN) % v͝d7_-{h`|CÅMשqdȃHxΉp5a_, 79rĦ-Dfqd2n,$d }F(&pd+1 7:ڢ 19V1X4FatE=E=ls|klP[ =0ʧFdxڠJ{1gf#Y1v~lE A/fO+k=D7A,g2ѨPR'T+DJ<S~U96VdǓ rQ[ S[CO)$M <0;rF@dMׇ֖Bk,Ue%k~geygI=9xK#4sYYsiF^N=#KHW3R?>#:"Smg%7,ʘ9y,g*2t3isVMAb>;#XǶ6~HvlJ̜в5||#;#^XHpՈƧ#IFSz^öt8i"CԈ&jH eeߜO(UJszEرXKbW3F^,ƉFr{VL#:Oռg. RW)ǗKVcLL!ES֐K9GAhpH{JmntI od#mdtGH";u8!P|aC l޸y{>fKWV\g$ʳ endstream endobj 60 0 obj 8651 endobj 61 0 obj (Microsoft Word - SAMBLASTER_SupplementalV5.docx) endobj 62 0 obj (Mac OS X 10.7.5 Quartz PDFContext) endobj 63 0 obj (gf4ea) endobj 64 0 obj (Word) endobj 65 0 obj (D:20141223180052Z00'00') endobj 66 0 obj () endobj 67 0 obj [ ] endobj 1 0 obj << /Title 61 0 R /Author 63 0 R /Producer 62 0 R /Creator 64 0 R /CreationDate 65 0 R /ModDate 65 0 R /Keywords 66 0 R /AAPL:Keywords 67 0 R >> endobj xref 0 68 0000000000 65535 f 0000108306 00000 n 0000011110 00000 n 0000030702 00000 n 0000000022 00000 n 0000011089 00000 n 0000011214 00000 n 0000014130 00000 n 0000030842 00000 n 0000036196 00000 n 0000000000 00000 n 0000042160 00000 n 0000000000 00000 n 0000088795 00000 n 0000000000 00000 n 0000075745 00000 n 0000015337 00000 n 0000000000 00000 n 0000098367 00000 n 0000011394 00000 n 0000014109 00000 n 0000014166 00000 n 0000015316 00000 n 0000030400 00000 n 0000015374 00000 n 0000030378 00000 n 0000030507 00000 n 0000000000 00000 n 0000061112 00000 n 0000030792 00000 n 0000031301 00000 n 0000031563 00000 n 0000036175 00000 n 0000036418 00000 n 0000036674 00000 n 0000042139 00000 n 0000043350 00000 n 0000042604 00000 n 0000043330 00000 n 0000043588 00000 n 0000061090 00000 n 0000061955 00000 n 0000061407 00000 n 0000061935 00000 n 0000062190 00000 n 0000075723 00000 n 0000076713 00000 n 0000076097 00000 n 0000076693 00000 n 0000076955 00000 n 0000088773 00000 n 0000089586 00000 n 0000089073 00000 n 0000089566 00000 n 0000089832 00000 n 0000098346 00000 n 0000099057 00000 n 0000098607 00000 n 0000099037 00000 n 0000099297 00000 n 0000108039 00000 n 0000108060 00000 n 0000108126 00000 n 0000108178 00000 n 0000108202 00000 n 0000108225 00000 n 0000108267 00000 n 0000108286 00000 n trailer << /Size 68 /Root 29 0 R /Info 1 0 R /ID [ <88994f60ef2871955da00be86e3ca1c7> <88994f60ef2871955da00be86e3ca1c7> ] >> startxref 108465 %%EOF samblaster-v.0.1.26/samblaster.cpp000066400000000000000000002155231366622051700170620ustar00rootroot00000000000000/* -*- mode: C++ ; indent-tabs-mode: nil ; c-file-style: "stroustrup" -*- Project: samblaster Fast mark duplicates in read-ID grouped SAM file. Also, optionally pull discordants, splitters, and/or unmappend/clipped reads. Author: Greg Faust (gf4ea@virginia.edu) Date: October 2013 Cite: SAMBLASTER: fast duplicate marking and structural variant read extraction GG Faust, IM Hall Bioinformatics 30 (17), 2503-2505 https://academic.oup.com/bioinformatics/article/30/17/2503/2748175 File: samblaster.cpp code file for the main routine and most of the other code. License Information: Copyright 2013-2020 Gregory G. Faust Licensed under the MIT license (the "License"); You may not use this file except in compliance with the License. You may obtain a copy of the License at http://opensource.org/licenses/MIT */ // This define is needed for portable definition of PRIu64 #define __STDC_FORMAT_MACROS #include #include #include #include #include #include #include "sbhash.h" // Rename common integer types. // I like having these shorter name. typedef uint64_t UINT64; typedef uint32_t UINT32; typedef int32_t INT32; // Some helper routines. // mempcpy is a GNU extension and not available everywhere. #ifndef _GNU_SOURCE inline void *mempcpy(void *dest, const void *src, size_t n) { return (char*) memcpy(dest, src, n) + n; } #endif inline bool streq(const char * s1, const char * s2) __attribute__((always_inline)); inline bool streq(const char * s1, const char * s2) { return (strcmp(s1, s2) == 0); } inline bool substr_of(const char * s1, const char * s2) __attribute__((always_inline)); inline bool substr_of(const char * s1, const char * s2) { return (strstr(s1, s2) != NULL); } // Declare error handling routines defined below. void fatalError(const char * errorStr); void fsError(const char * filename); /////////////////////////////////////////////////////////////////////////////// // Runtime Statistics /////////////////////////////////////////////////////////////////////////////// // Stuff needed for timings. // To turn timing off, set the below to 0. #define TIMING 1 #if TIMING // A convenience function for outputing time is seconds in a more useful metric. void fprintTimeSeconds (FILE * out, double seconds, int precision) { double totalseconds = seconds; int hours = seconds/3600.; if (hours > 0) { seconds -= hours * 3600; fprintf(out, "%dH", hours); } int minutes = seconds/60.; if (minutes > 0) { seconds -= minutes * 60; fprintf(out, "%dM", minutes); } if (hours + minutes > 0) { fprintf(out, "%.0fS", seconds); fprintf(out, "(%.*fS)", precision, totalseconds); } else fprintf(out, "%.*fS", precision, totalseconds); } void fprintTimeMicroSeconds (FILE * out, UINT64 microSeconds, int precision) { fprintTimeSeconds(out, ((double)microSeconds/1000000.0), precision); } inline UINT64 diffTVs (struct timeval * startTV, struct timeval * endTV) { return (((endTV->tv_sec - startTV->tv_sec) * 1000000) + (endTV->tv_usec - startTV->tv_usec)); } #include #include #include #endif // TIMING /////////////////////////////////////////////////////////////////////////////// // Split Lines /////////////////////////////////////////////////////////////////////////////// // The structure to store "split" input lines, especially SAM lines. // They form a singly linked list so that we can form groups of them, // and also so that we can keep a freelist of them. // Reference offsets (pos) can be negative or run off the end of a contig due to clipping. // Therefore, we will use a padding strategy. // The space allocated to each contig will be padded by twice the max read length. // This leaves room for both offset underflow and overflow. // And all offsets will be shifted higher by the max read length. // This will eliminate negative offsets and "center" offsets within the offset range for the contig. typedef INT32 pos_t; // We need to pre-define these for the SAM specific fields. typedef UINT64 sgn_t; // Type for signatures for offsets and lengths. // And the type itself for the next pointer. typedef struct splitLine splitLine_t; splitLine_t * splitLineFreeList = NULL; struct splitLine { // General fields for any split line. splitLine_t * next; char * buffer; int bufLen; size_t maxBufLen; char **fields; int numFields; int maxFields; bool split; // Special SAM fields that we need to access as other than strings. // It this were a class, these would be in a subclass. int flag; pos_t pos; int seqNum; pos_t binPos; int binNum; int SQO; int EQO; int sclip; int eclip; int rapos; int raLen; int qaLen; bool CIGARprocessed; bool discordant; bool splitter; bool unmappedClipped; }; // Creator for splitLine splitLine_t * makeSplitLine() { splitLine_t * line = (splitLine_t *)malloc(sizeof(splitLine_t)); line->bufLen = 0; line->maxBufLen = 1000; line->buffer = (char *)malloc(line->maxBufLen+1); line->numFields = 0; line->maxFields = 100; line->fields = (char **)malloc(line->maxFields * sizeof(char *)); return line; } // Destructor for split line. void deleteSplitLine(splitLine_t * line) { free(line->buffer); free(line->fields); free(line); } // Descructor for a list of splitLines void cleanUpSplitLines() { splitLine_t * l = splitLineFreeList; while (l != NULL) { splitLine_t * next = l->next; deleteSplitLine(l); l = next; } } // Like descructor for splitLine except don't free memory. // Instead, put the linked list of objects back on the free list. void disposeSplitLines(splitLine_t * line) { // First find the last line in the list. // Then get rid of them all. splitLine_t * last = line; for (splitLine_t * l = line->next; l!=NULL; l = l->next) last = l; last->next = splitLineFreeList; splitLineFreeList = line; } // Like constuctor, except take struct off free list if there is one. splitLine_t * getSplitLine() { splitLine_t * line; if (splitLineFreeList == NULL) { line = makeSplitLine(); } else { line = splitLineFreeList; splitLineFreeList = splitLineFreeList->next; } line->next = NULL; line->CIGARprocessed = false; // Mark these here so that the code for these doesn't have to stand on its head to do it. line->discordant = false; line->splitter = false; line->unmappedClipped = false; return line; } // Split the line into fields. void splitSplitLine(splitLine_t * line, int maxSplits) { line->numFields = 0; int fieldStart = 0; // replace the newline with a tab so that it works like the rest of the fields. line->buffer[line->bufLen-1] = '\t'; for (int i=0; ibufLen; ++i) { if (line->buffer[i] == '\t') { line->fields[line->numFields] = line->buffer + fieldStart; line->numFields += 1; if (line->numFields == maxSplits) break; line->buffer[i] = 0; // Get ready for the next iteration. fieldStart = i+1; } } // replace the tab at the end of the line with a null char to terminate the final string. line->buffer[line->bufLen-1] = 0; line->split = true; } // Unsplit the fields back into a single string. // This will mung the strings, so only call this when all processing on the line is done. void unsplitSplitLine(splitLine_t * line) { // First make sure we are still split. if (!line->split) return; // First undo the splits. // We will undo the splits backwards from the next field to avoid having to calculate strlen each time. for (int i=1; inumFields; ++i) { line->fields[i][-1] = '\t'; } // Now put the newline back in. line->buffer[line->bufLen-1] = '\n'; // Mark as no longer split. line->split = false; } // Resize the buffer of a splitLine. // Since the new buffer may not be in the same place, we need to first unsplit, resize, then resplit. void resizeSplitLine(splitLine_t * line, int newsize) { // First unsplit it. unsplitSplitLine(line); // Resize the buffer, giving a little extra room. line->maxBufLen = newsize + 50; line->buffer = (char *)realloc(line->buffer, line->maxBufLen); if (line->buffer == NULL) { fatalError("samblaster: Failed to reallocate to a larger read buffer size.\n"); } // Now resplit the line. splitSplitLine(line, line->numFields); } // Change a field into a value. // This will be tough given how we output lines. // So, we might have to try a few things. // Start with simply expanding/contracting the string to put in the new value. void changeFieldSplitLine(splitLine_t * line, int fnum, char * newValue) { // What we do will depend on the lengths of the two strings. // So, start by calculaing these only once. char * fp = line->fields[fnum]; int oldLen = strlen(fp); int newLen = strlen(newValue); // Now see if we need to first change the length of the whole line. int move = newLen - oldLen; if (move != 0) { // This should never happen, but to be robust we need to check. // It is messy to fix it, as all the field ptrs will now be wrong. if ((size_t)(line->bufLen + move) >= line->maxBufLen) { resizeSplitLine(line, line->bufLen + move); fp = line->fields[fnum]; } // Calculate the size of the tail that is still needed. int distance = 1 + line->bufLen - (fp - line->buffer) - oldLen; // Do the copy. memmove(fp+newLen, fp+oldLen, distance); // Correct the total length of the buffer. line->bufLen += move; // We need to correct the other ptrs as well. for (int i=fnum+1; inumFields; i++) line->fields[i] += move; } // Copy in the new value. memcpy(fp, newValue, newLen); } void addTag(splitLine_t * line, const char * header, const char * val) { int hl = strlen(header); int vl = strlen(val); // Make sure everything will fit. int newlen = line->bufLen + hl + vl; if ((size_t)newlen >= line->maxBufLen) { resizeSplitLine(line, newlen); } // Copy over the header and the value. char * ptr = line->buffer + line->bufLen - 1; ptr = (char *)mempcpy(ptr, header, hl); ptr = (char *)mempcpy(ptr, val, vl); // Add the null terminator for the field, and for the record. ptr[0] = 0; ptr[1] = 0; // Fix the buffer length. line->bufLen = newlen; } #if 0 // This was originally written to test for mismatched MC tags between BWA MEM and samblaster. // I keep it around, as some version of this would be useful if/when we add support for RGs and/or UMIs. // Note that this conses up a string, so the caller will have to free it later to avoid a leak. char * getTagVal (char * tags, char * tagID) { char * startptr = NULL; char * endptr = NULL; char * retval = NULL; // Find the start and end of the tag field startptr = strstr(tags, tagID); if (startptr == NULL) return strdup(""); endptr = strchr(startptr, '\t'); if (endptr == NULL) endptr = strchr(startptr, 0); // temporarily put a null in for the delimiter char save = endptr [0]; endptr [0] = 0; retval = strdup (startptr+strlen(tagID)); endptr [0] = save; // Return the dupped string return retval; } #endif // Read a line from the file and split it. splitLine_t * readLine(FILE * input) { splitLine_t * sline = getSplitLine(); sline->bufLen = getline(&sline->buffer, &sline->maxBufLen, input); if (sline->bufLen < 1) { disposeSplitLines(sline); return NULL; } splitSplitLine(sline, 12); return sline; } inline void outputString(char * str, FILE * output) { // Do the error checking here so we don't have to do it elsewhere. if (fputs(str, output) < 0) { fatalError("samblaster: Unable to write to output file.\n"); } } // Output the line. inline void writeLine(splitLine_t * line, FILE * output) { unsplitSplitLine(line); outputString(line->buffer, output); } // Check the first line of a file (e.g. input) for bam signature. void checkBAMfile(splitLine_t * line) { // If the file is a bam file, we can't rely on fields. // So, look at the underlying buffer for the line. // First define the signature to look for. int values[] = {31, -117, 8, 4, 66, 67, 2}; int offsets[] = {0, 1, 2, 3, 12, 13, 14}; int count = 7; // Check for empty file or likely a sam file with a header. // This is necessary, as the @HD line may not be long enough to check for the BAM signature. if (line == NULL) fatalError("samblaster: Input file is empty. Exiting.\n"); if (line->buffer[0] == '@') return; // If a SAM file has no header, an alignment row should easily be long enough. if (line->bufLen <= offsets[count-1]) fatalError("samblaster: Input file is empty. Exiting.\n"); // Check for the BAM signature. for (int i=0; ibuffer[offsets[i]] != values[i]) return; } // If we are here, we almost certainly have a bamfile. fatalError("samblaster: Input file appears to be in BAM format. SAM input is required. Exiting.\n"); } /////////////////////////////////////////////////////////////////////////////// // SAM and signature set related structures. /////////////////////////////////////////////////////////////////////////////// // Define SAM field offsets. #define QNAME 0 #define FLAG 1 #define RNAME 2 #define POS 3 #define MAPQ 4 #define CIGAR 5 #define RNEXT 6 #define PNEXT 7 #define TLEN 8 #define SEQ 9 #define QUAL 10 #define TAGS 11 // Define SAM flag accessors. #define MULTI_SEGS 0x1 #define FIRST_SEG 0x40 #define SECOND_SEG 0x80 inline bool checkFlag(splitLine_t * line, int bits) { return ((line->flag & bits) != 0); } inline void setFlag(splitLine_t * line, int bits) { line->flag |= bits; } inline bool isPaired(splitLine_t * line) { return checkFlag(line, MULTI_SEGS); } inline bool isConcordant(splitLine_t * line) { return checkFlag(line, 0x2); } inline bool isDiscordant(splitLine_t * line) { return !isConcordant(line); } inline bool isUnmapped(splitLine_t * line) { return checkFlag(line, 0x4); } inline bool isNextUnmapped(splitLine_t * line) { return checkFlag(line, 0x8); } inline bool isNextMapped(splitLine_t * line) { return !isNextUnmapped(line); } inline bool isMapped(splitLine_t * line) { return !isUnmapped(line); } inline bool isReverseStrand(splitLine_t * line) { return checkFlag(line, 0x10); } inline bool isForwardStrand(splitLine_t * line) { return !isReverseStrand(line); } inline bool isFirstRead(splitLine_t * line) { return checkFlag(line, FIRST_SEG); } inline bool isSecondRead(splitLine_t * line) { return checkFlag(line, SECOND_SEG); } // These determine alignment type. // Things may get more complicated than this once we have alternate contigs such as in build 38 of human genome. inline bool isPrimaryAlignment(splitLine_t * line) { return !(checkFlag(line, 0x100) || checkFlag(line, 0x800)); } // We have to hande secondard and complementary alignments differently depending on compatMode. // So, we store which bits are being included in each. int complementaryBits = 0x800; inline bool isComplementaryAlignment(splitLine_t * line) { return checkFlag(line, complementaryBits); } int secondaryBits = 0x100; inline bool isSecondaryAlignment(splitLine_t * line) { return checkFlag(line, secondaryBits); } inline bool isDuplicate(splitLine_t * line) { return checkFlag(line, 0x400); } inline void setDuplicate(splitLine_t * line) { setFlag(line, 0x400); } typedef hashTable_t sigSet_t; inline int str2int (char * str) { return strtol(str, NULL, 0); } // Need to change this if pos is unsigned. inline pos_t str2pos (char * str) { return strtol(str, NULL, 0); } // Temp buffer to use to form new flag field when marking dups. char tempBuf[10]; inline void markDup(splitLine_t * line) { setDuplicate(line); sprintf(tempBuf, "%d", line->flag); changeFieldSplitLine(line, FLAG, tempBuf); } // Special version of write line that appends an id number to the output. // Used to output splitters. void writeSAMlineWithIdNum(splitLine_t * line, FILE * output) { // Unsplit the line. unsplitSplitLine(line); // Split it two ways to isolate the id field. splitSplitLine(line, 2); outputString(line->fields[0], output); if (isPaired(line)) fprintf(output, "_%d\t", isFirstRead(line) ? 1 : 2); else fprintf(output, "\t"); outputString(line->fields[1], output); fprintf(output, "\n"); } /////////////////////////////////////////////////////////////////////////////// // Sequence Map /////////////////////////////////////////////////////////////////////////////// // We use a map instead of a hash map for sequence names. // This is because the default hash function on char * hashes the ptr values. // So, we would need to define our own hash on char * to get things to work properly. // Not worth it for a structure holding so few members. // Function needed to get char * map to work. struct less_str { bool operator()(char const *a, char const *b) const { return strcmp(a, b) < 0; } }; // This stores the map between sequence names and sequence numbers. typedef std::map seqMap_t; inline void addSeq(seqMap_t * seqs, char * item, int val) { (*seqs)[item] = val; } /////////////////////////////////////////////////////////////////////////////// // Struct for processing state /////////////////////////////////////////////////////////////////////////////// struct state_struct { char * inputFileName; FILE * inputFile; char * outputFileName; FILE * outputFile; FILE * discordantFile; char * discordantFileName; FILE * splitterFile; char * splitterFileName; FILE * unmappedClippedFile; char * unmappedClippedFileName; sigSet_t * sigs; seqMap_t seqs; UINT32 * seqLens; UINT64 * seqOffs; splitLine_t ** splitterArray; int splitterArrayMaxSize; UINT32 sigArraySize; int binCount; int minNonOverlap; int maxReadLength; int maxSplitCount; int minIndelSize; int maxUnmappedBases; int minClip; int unmappedFastq; bool acceptDups; bool excludeDups; bool removeDups; bool addMateTags; bool compatMode; bool ignoreUnmated; bool quiet; }; typedef struct state_struct state_t; state_t * makeState () { state_t * s = new state_t(); s->inputFile = stdin; s->inputFileName = (char *)"stdin"; s->outputFile = stdout; s->outputFileName = (char *)"stdout"; s->discordantFile = NULL; s->discordantFileName = (char *)""; s->splitterFile = NULL; s->splitterFileName = (char *)""; s->unmappedClippedFile = NULL; s->unmappedClippedFileName = (char *)""; s->sigs = NULL; s->minNonOverlap = 20; s->maxSplitCount = 2; s->minIndelSize = 50; s->maxUnmappedBases = 50; s->minClip = 20; s->maxReadLength = 500; s->acceptDups = false; s->excludeDups = false; s->removeDups = false; s->addMateTags = false; s->compatMode = false; s->ignoreUnmated = false; s->quiet = false; // Start this as -1 to indicate we don't know yet. // Once we are outputting our first line, we will decide. s->unmappedFastq = -1; // Used as a temporary location for ptrs to splitter for sort routine. s->splitterArrayMaxSize = 1000; s->splitterArray = (splitLine_t **)(malloc(s->splitterArrayMaxSize * sizeof(splitLine_t *))); return s; } void deleteState(state_t * s) { free(s->splitterArray); if (s->sigs != NULL) { // delete[] s->sigs; for (UINT32 i=0; isigArraySize; i++) deleteHashTable(&(s->sigs[i])); free (s->sigs); } for (seqMap_t::iterator iter = s->seqs.begin(); iter != s->seqs.end(); ++iter) { free((char *)(iter->first)); } if (s->seqLens != NULL) free(s->seqLens); if (s->seqOffs != NULL) free(s->seqOffs); delete s; } /////////////////////////////////////////////////////////////////////////////// // Signatures /////////////////////////////////////////////////////////////////////////////// // We now calculate signatures as offsets into a super contig that includes the entire genome. // And partition it into equaly sized bins. // This performs better on genomes with large numbers of small contigs, // without performance degradation on more standard genomes. // Thanks to https://github.com/carsonhh for the suggestion. ///////////////////////////////////////////////////////////////////////////// template inline sgn_t calcSig(splitLine_t * first, splitLine_t * second) { UINT64 final; if (orphan) { // For an orphan, we only use information fron the second read. final = second->binPos; } else { // Total nonsense to get the compiler to actually work. UINT64 t1 = first->binPos; UINT64 t2 = t1 << 32; final = t2 | second->binPos; } return (sgn_t)final; } template inline UINT32 calcSigArrOff(splitLine_t * first, splitLine_t * second, int binCount) { UINT32 s1, s2; if (orphan) { // For orphans, we only use the binNum of the second, and treat it as if on the forward strand. s1 = 0; s2 = (second->binNum * 2); } else { s1 = (first->binNum * 2) + (isReverseStrand(first) ? 1 : 0); s2 = (second->binNum * 2) + (isReverseStrand(second) ? 1 : 0); } return (UINT32)(s1 * binCount * 2) + s2; } /////////////////////////////////////////////////////////////////////////////// // Sequences /////////////////////////////////////////////////////////////////////////////// inline int getSeqNum(splitLine_t * line, int field, state_t * state) __attribute__((always_inline)); inline int getSeqNum(splitLine_t * line, int field, state_t * state) { seqMap_t::iterator findret = state->seqs.find(line->fields[field]); if (findret == state->seqs.end()) { char * temp; asprintf(&temp, "samblaster: Unable to find sequence '%s' in sequence map for readid %s\n", line->fields[field], line->fields[QNAME]); fatalError(temp); } return findret->second; } /////////////////////////////////////////////////////////////////////////////// // Helpers to process CIGAR strings /////////////////////////////////////////////////////////////////////////////// // This will parse a base 10 int, and change ptr to one char beyond the end of the number. inline int parseNextInt(char **ptr) { int num = 0; for (char curChar = (*ptr)[0]; curChar != 0; curChar = (++(*ptr))[0]) { int digit = curChar - '0'; if (digit >= 0 && digit <= 9) num = num*10 + digit; else break; } return num; } // This will the current char, and move the ptr ahead by one. inline char parseNextOpCode(char **ptr) { return ((*ptr)++)[0]; } // This just test for end of string. inline bool moreCigarOps(char *ptr) { return (ptr[0] != 0); } void processCIGAR(splitLine_t * line) { if (line->CIGARprocessed) return; char * cigar = line->fields[CIGAR]; line->raLen = 0; line->qaLen = 0; line->sclip = 0; line->eclip = 0; bool first = true; while (moreCigarOps(cigar)) { int opLen = parseNextInt(&cigar); char opCode = parseNextOpCode(&cigar); if (opCode == 'M' || opCode == '=' || opCode == 'X') { line->raLen += opLen; line->qaLen += opLen; first = false; } else if (opCode == 'S' || opCode == 'H') { if (first) line->sclip += opLen; else line->eclip += opLen; } else if (opCode == 'D' || opCode == 'N') { line->raLen += opLen; } else if (opCode == 'I') { line->qaLen += opLen; } else { fprintf(stderr, "Unknown opcode '%c' in CIGAR string: '%s'\n", opCode, line->fields[CIGAR]); } } line->rapos = str2pos(line->fields[POS]); if (isForwardStrand(line)) { line->pos = line->rapos - line->sclip; line->SQO = line->sclip; line->EQO = line->sclip + line->qaLen - 1; } else { line->pos = line->rapos + line->raLen + line->eclip - 1; line->SQO = line->eclip; line->EQO = line->eclip + line->qaLen - 1; } line->CIGARprocessed = true; } inline int getStartDiag(splitLine_t * line) { // SRO - SQO (not strand normalized) // Simplify the following. // return (str2pos(line->fields[POS])) - line->sclip; return line->rapos - line->sclip; } inline int getEndDiag(splitLine_t * line) { // ERO - EQO (not strand normalized) // Simplify the following // return (line->rapos + line->raLen - 1) - (line->sclip + line->qaLen - 1) return (line->rapos + line->raLen) - (line->sclip + line->qaLen); } /////////////////////////////////////////////////////////////////////////////// // Process SAM Blocks /////////////////////////////////////////////////////////////////////////////// // This was historically 27 bits as that was large enough to hold any human chrom 1 offset. // Now that we are using a synthetic genome representation, 27 has no special meaning. // There are interesting space/time trade-offs between how large we make the // synthetic chroms and how large the various hash tables become. // If/when we handle RGs and/or UMIs, the signature scheme will change anyway. // Therefore, leave at 27 for now to match space/time tradeoffs of earlier releases. #define BIN_SHIFT ((UINT64)27) //bin window is 27 bits wide #define BIN_MASK ((UINT64)((1 << BIN_SHIFT)-1)) //bin window is 27 bits wide // This is apparently no longer called. void outputSAMBlock(splitLine_t * block, FILE * output) { for (splitLine_t * line = block; line != NULL; line = line->next) { writeLine(line, output); } disposeSplitLines(block); } inline bool needSwap(splitLine_t * first, splitLine_t * second) { // Sort first by ref offset. if (first->pos > second->pos) return true; if (first->pos < second->pos) return false; // Now by seq number. if (first->seqNum > second->seqNum) return true; if (first->seqNum < second->seqNum) return false; // Now by strand. // If they are both the same strand, it makes no difference which on is first. if (isReverseStrand(first) == isReverseStrand(second)) return false; if (isReverseStrand(first) && isForwardStrand(second)) return true; return false; } inline void swapPtrs(splitLine_t ** first, splitLine_t ** second) { splitLine_t * temp = *first; *first = *second; *second = temp; } template int fillTempLineArray(splitLine_t * block, state_t * state, int mask, bool flagValue); void addMTs(splitLine_t * block, state_t * state, splitLine_t * first, splitLine_t * second) { if (!isMapped(first)) return; int mask = (FIRST_SEG | SECOND_SEG); // Get the list of reads that match the other read of the pair. int count = fillTempLineArray(block, state, second->flag & mask, true); for (int i=0; isplitterArray[i]; if (!substr_of(line->fields[TAGS], "MC:Z:")) addTag(line, " MC:Z:", first->fields[CIGAR]); if (!substr_of(line->fields[TAGS], "MQ:i:")) addTag(line, " MQ:i:", first->fields[MAPQ]); } } void brokenBlock(splitLine_t *block, int count) { char * temp; asprintf(&temp, "samblaster: Can't find first and/or second of pair in sam block of length %d for id: %s\n%s%s:%s\n%s", count, block->fields[QNAME], "samblaster: At location: ", block->fields[RNAME], block->fields[POS], "samblaster: Are you sure the input is sorted by read ids?"); fatalError(temp); } // Some fields for statistics. // In order to form a partition of all the reads (i.e. account for all ids) // we need weird categories such as noPrimary, and unmapped orphans. UINT64 idCount = 0; UINT64 dupCount = 0; UINT64 noPrimaryIdCount = 0; UINT64 bothUnmappedIdCount = 0; UINT64 unmappedOrphanIdCount = 0; UINT64 mappedOrphanIdCount = 0; UINT64 orphanDupCount = 0; UINT64 bothMappedDupCount = 0; UINT64 bothMappedIdCount = 0; UINT64 discCount = 0; UINT64 splitCount = 0; UINT64 unmapClipCount = 0; UINT64 unmatedCount = 0; UINT64 readTooLongCount = 0; INT32 readTooLongMax = 0; // We also make the state (and timing info) global to aid error processing state_t * state; #ifdef TIMING time_t startTime; struct timeval startRUTime; #endif // The routines to do implement our padding strategy, as needed directly below. inline int padLength(int length, state_t * state) __attribute__((always_inline)); inline int padLength(int length, state_t * state) { return length + (2 * state->maxReadLength); } inline int padPos(int pos, state_t * state) __attribute__((always_inline)); inline int padPos(int pos, state_t * state) { return pos + state->maxReadLength; } // This calculates the position of the read in the sequence array, and in the reference genome. void prepareSigValues(splitLine_t * line, state_t * state, bool orphan) { // Calculate and store the reference sequence name and sequence relative position. processCIGAR(line); // Calculate the full query length, which is the aligned length plus clips. int fullqlen = (line->sclip + line->qaLen + line->eclip); // Make sure we don't have any problems with this read longer than our padding will handle. if (fullqlen > state->maxReadLength) { // The read is longer than our padding will handle. readTooLongCount += 1; if (fullqlen > readTooLongMax) readTooLongMax = fullqlen; // We should only ever miss the bins entirely if the read is longer than maxReadLength due to our padding of the reference. if (line->binNum < 0 || line->binNum > state->binCount) { char * temp; asprintf(&temp, "samblaster: Calculated read position falls outside the reference for read id: %s\n" "samblaster: Consider rerunning with higher --maxReadLength.\n", line->fields[QNAME]); fatalError(temp); } } if (orphan && isReverseStrand(line)) { // We want to treat this read as if it is on the forward strand. line->pos = line->rapos - line->sclip; } // Now get the sequence number line->seqNum = getSeqNum(line, RNAME, state); // Calculate the genome relative position // We need to do the read pos padding to handle negative calculated sequence position for the alignment. UINT64 seqOff = state->seqOffs[line->seqNum]; int paddedPos = padPos(line->pos, state); line->binNum = (seqOff + paddedPos) >> BIN_SHIFT; line->binPos = (seqOff + paddedPos) & BIN_MASK; } // This is the main workhorse that determines if lines are dups or not. void markDupsDiscordants(splitLine_t * block, state_t * state) { splitLine_t * first = NULL; splitLine_t * second = NULL; int count = 0; for (splitLine_t * line = block; line != NULL; line = line->next) { count += 1; // Do this conversion once and store the result. line->flag = str2int(line->fields[FLAG]); // We make our duplicate decisions based solely on primary alignments. if (!isPrimaryAlignment(line)) continue; // Allow unpaired reads to go through (as the second so that signature is correct). // According to the SAM spec, this must be checked first. if (!isPaired(line)) second = line; // Figure out if this is the first half or second half of a pair. else if (isFirstRead(line)) first = line; else if (isSecondRead(line)) second = line; } // Figure out what type of "pair" we have. // First get rid of the useless case of having no first AND no second. if (first == NULL && second == NULL) { if (state->ignoreUnmated) { noPrimaryIdCount += 1; return; } brokenBlock(block, count); } // Now see if we have orphan with the unmapped read missing. bool orphan = false; if (first == NULL || second == NULL) { // Get the NULL one in the first slot. if (second == NULL) swapPtrs(&first, &second); // If the only read says its paired, and it is unmapped or its mate is mapped, something is wrong. if (isPaired(second) && (isUnmapped(second) || isNextMapped(second))) { if (state->ignoreUnmated) { unmatedCount += 1; return; } brokenBlock(block, count); } // If the only read we have is unmapped, then it can't be a dup. if (isUnmapped(second)) { if (state->ignoreUnmated) { unmappedOrphanIdCount += 1; return; } brokenBlock(block, count); } // We have a processable singleton. orphan = true; } else { // Handle the addition of MC and MQ tags if requested. if (state->addMateTags) { addMTs(block, state, first, second); addMTs(block, state, second, first); } // Never mark pairs as dups if both sides are unmapped. if (isUnmapped(first) && isUnmapped(second)) { bothUnmappedIdCount += 1; return; } // We need to properly handle orphans to get the correct reference offsets and sequence numbers. orphan = (isUnmapped(first) || isUnmapped(second)); // Orphan that needs to be swapped. // We need the unmapped one in the first slot so that they won't all collide in the hash table. if (isMapped(first) && isUnmapped(second)) { swapPtrs(&first, &second); } } // Keep track of the ID category. if (orphan) mappedOrphanIdCount += 1; else bothMappedIdCount +=1; // Now look for duplicates. if (!state->acceptDups) { // Calculate and store the second position and sequence name. prepareSigValues(second, state, orphan); sgn_t sig; UINT32 off; if (orphan) { // Now find the signature of the pair. sig = calcSig(first, second); // Calculate the offset into the signatures array. off = calcSigArrOff(first, second, state->binCount); } else { // Not an orphan, so handle first on its own. prepareSigValues(first, state, false); // The fact of which alignment is first or second in the template is not relevant for determining dups. // Therefore, we normalize the pairs based on their characteristics. // We have already swapped orphans. // Otherwise, sort by pos, and if equal, sort by sequence num, then by strand. if (needSwap(first, second)) swapPtrs(&first, &second); // Now find the signature of the pair. sig = calcSig(first, second); // Calculate the offset into the signatures array. off = calcSigArrOff(first, second, state->binCount); } // Attempt insert into the sigs structure. // The return value will tell us if it was already there. bool insert = hashTableInsert(&(state->sigs[off]), sig); #ifdef DEBUG // This debug output allows the discovery of all members of a duplicate group including the read not marked as a duplicate. // This could otherwise not be figured out from the output sam file. fprintf(stderr, "%" PRIu64 "\t%s\t%6d\t%s\t%" PRIu64 "\t%s\n", idCount, second->fields[QNAME], second->flag, insert ? "First" : "Dup", sig, second->fields[RNAME]); #endif // Check if the insertion actually happened. if (!insert) { dupCount += 1; if (orphan) orphanDupCount += 1; else bothMappedDupCount += 1; // We always mark all or none of a block as dup. for (splitLine_t * line = block; line != NULL; line = line->next) { markDup(line); } } } // The first and second help us mark the discordants. // Both sides mapped, but pair not properly aligned. if (!orphan && isDiscordant(first)) { first->discordant = true; second->discordant = true; } } // Sort ascending in SQO. int compQOs(const void * p1, const void * p2) { splitLine_t * l1 = (*(splitLine_t **)p1); splitLine_t * l2 = (*(splitLine_t **)p2); return (l1->SQO - l2->SQO); } template int fillTempLineArray(splitLine_t * block, state_t * state, int mask, bool flagValue) { // Count the secondaries we have for this read (if any), and store their ptrs into an array. int count = 0; for (splitLine_t * line = block; line != NULL; line = line->next) { // For all the ones that are the current read of interest.... // Check if they are a primary or complementary alignment. if (checkFlag(line, mask) == flagValue && !(excludeSecondaries && isSecondaryAlignment(line))) { // Add the ptr to this line to the sort array. // If it won't fit, double the array size. if (count >= state->splitterArrayMaxSize) { state->splitterArrayMaxSize *= 2; state->splitterArray = (splitLine_t **)(realloc(state->splitterArray, state->splitterArrayMaxSize * sizeof(splitLine_t *))); } state->splitterArray[count] = line; count += 1; } } return count; } void markSplitterUnmappedClipped(splitLine_t * block, state_t * state, int mask, bool flagValue) { // Count the secondaries we have for this read (if any), and store their ptrs into an array. int count = fillTempLineArray(block, state, mask, flagValue); // We have the lines of interest in an array. // Decide what to do next based on the number of reads. if (count == 0) return; if (count == 1) { if (state->unmappedClippedFile == NULL) return; // Process unmapped or clipped. splitLine_t * line = state->splitterArray[0]; // Unmapped or clipped alignments should be primary. if (!isPrimaryAlignment(line)) return; if (isUnmapped(line)) { line->unmappedClipped = true; } else { // Process the CIGAR string. // As this is expensive, we delay as long as possible. processCIGAR(line); if (line->sclip >= state->minClip || line->eclip >= state->minClip) { line->unmappedClipped = true; } } return; } // See if we need to process for splitters. if (state->splitterFile == NULL || count > state->maxSplitCount) return; // Calculate the query positions (for sorting) and do other preprocessing. for (int i=0; isplitterArray[i]; // Make sure the primary is mapped! if (isPrimaryAlignment(line) && isUnmapped(line)) return; processCIGAR(line); } // We need to sort it by strand normalized query offsets. qsort(state->splitterArray, count, sizeof(splitLine_t *), compQOs); // Now check for pairs that match the desired parameters. for (int i=1; isplitterArray[i-1]; splitLine_t * right = state->splitterArray[i]; // First check for minNonOverlap. // We don't allow negative overlap, as that will lead to wrong non-overlap calculation. int overlap = std::max(1 + std::min(left->EQO, right->EQO) - std::max(left->SQO, right->SQO), 0); int alen1 = 1 + left->EQO - left->SQO; int alen2 = 1 + right->EQO - right->SQO; int mno = std::min(alen1-overlap, alen2-overlap); if (mno < state->minNonOverlap) continue; // Now check for the deserts and diagonal difference. // If they are on different chroms or strands, they pass without the other checks. // Since we only care if the sequences are the same, we don't need seqNums. // Instead just compare the strings! if (streq(left->fields[RNAME], right->fields[RNAME]) && (isReverseStrand(left) == isReverseStrand(right))) { // The start and end diags might be different if there is an indel in the alignments. // So, we use the end on the left, and the start on the right. // This will give us the net diag difference between them. int leftDiag, rightDiag, insSize; if (isReverseStrand(left)) { leftDiag = getStartDiag(left); rightDiag = getEndDiag(right); insSize = rightDiag - leftDiag; } else { leftDiag = getEndDiag(left); rightDiag = getStartDiag(right); insSize = leftDiag - rightDiag; } int desert = right->SQO - left->EQO - 1; // The absolute value will handle both inserts and deletes. // So check indel is big enough, and that there are not too many unmapped bases. // Subtract the inadvertant desert gap of inserts. if ((abs(insSize) < state->minIndelSize) || ((desert > 0) && ((desert - (int)std::max(0, insSize)) > state->maxUnmappedBases))) continue; } // We made it through the gamet. // Mark this pair for output. left->splitter = true; right->splitter = true; } } void writeUnmappedClipped(splitLine_t * line, state_t * state) { // Check if we are outputting fasta or fastq. if (state->unmappedFastq == -1) state->unmappedFastq = (streq(line->fields[QUAL], "*") ? 0 : 1); // Print the first line. char firstChar = (state->unmappedFastq) ? '@' : '>'; if (isPaired(line)) fprintf(state->unmappedClippedFile, "%c%s_%d\n", firstChar, line->fields[QNAME], (isFirstRead(line) ? 1 : 2)); else fprintf(state->unmappedClippedFile, "%c%s\n", firstChar, line->fields[QNAME]); if (state->unmappedFastq) fprintf(state->unmappedClippedFile, "%s\n+\n%s\n", line->fields[SEQ], line->fields[QUAL]); else fprintf(state->unmappedClippedFile, "%s\n", line->fields[SEQ]); } void processSAMBlock(splitLine_t * block, state_t * state) { idCount += 1; // First mark dups and find the discordants. // These share a lot of looking at flag bits, so make sense to put together. markDupsDiscordants(block, state); // Look for splitters. // Since this is expensive, don't do it unless the user asked for them. if (state->splitterFile != NULL || state->unmappedClippedFile != NULL) { // Check the first read for splitter. markSplitterUnmappedClipped(block, state, FIRST_SEG, true); // Check the second read for splitter. markSplitterUnmappedClipped(block, state, SECOND_SEG, true); // Check for a singleton read markSplitterUnmappedClipped(block, state, MULTI_SEGS, false); } // Now do the output. for (splitLine_t * line = block; line != NULL; line = line->next) { // Do the unmapped file first, as it is not sam, and doesn't sew the line back together. if (state->unmappedClippedFile != NULL && line->unmappedClipped && !(state->excludeDups && isDuplicate(line))) { writeUnmappedClipped(line, state); unmapClipCount += 1; } // Write to the output file. if (!(state->removeDups && isDuplicate(line))) { writeLine(line, state->outputFile); } // Write to discordant file if appropriate. if (state->discordantFile != NULL && line->discordant && !(state->excludeDups && isDuplicate(line))) { writeLine(line, state->discordantFile); discCount += 1; } // Write to splitter file if appropriate. if (state->splitterFile != NULL && line->splitter && !(state->excludeDups && isDuplicate(line))) { writeSAMlineWithIdNum(line, state->splitterFile); splitCount += 1; } } disposeSplitLines(block); } /////////////////////////////////////////////////////////////////////////////// // Main Routine with helpers. /////////////////////////////////////////////////////////////////////////////// // Output the @PG line information into a SAM file header void printPGsamLine(FILE * f, state_t * s) { if (f == NULL) return; fprintf(f, "@PG\tID:SAMBLASTER\tVN:0.1.%d\tCL:samblaster -i %s -o %s", BUILDNUM, s->inputFileName, s->outputFileName); if (s->compatMode) fprintf(f, " -M"); if (s->acceptDups) fprintf(f, " --acceptDupMarks"); if (s->removeDups) fprintf(f, " --removeDups"); else if (s->excludeDups && (s->discordantFile != NULL || s->splitterFile != NULL || s->unmappedClippedFile != NULL)) fprintf(f, " --excludeDups"); if (s->addMateTags) fprintf(f, " --addMateTags"); if (s->ignoreUnmated) fprintf(f, " --ignoreUnmated"); if (s->maxReadLength != 500) fprintf(f, " --maxReadLength %d", s->maxReadLength); if (s->discordantFile != NULL) fprintf(f, " -d %s", s->discordantFileName); if (s->splitterFile != NULL) fprintf(f, " -s %s --maxSplitCount %d --maxUnmappedBases %d --minIndelSize %d --minNonOverlap %d", s->splitterFileName, s->maxSplitCount, s->maxUnmappedBases, s->minIndelSize, s->minNonOverlap); if (s->unmappedClippedFile != NULL) fprintf(f, " -u %s --minClipSize %d", s->unmappedClippedFileName, s->minClip); fprintf(f, "\n"); } void printVersionString() { fprintf(stderr, "samblaster: Version 0.1.%d\n", BUILDNUM); } void printUsageString() { const char* useString = "Author: Greg Faust (gf4ea@virginia.edu)\n" "Tool to mark duplicates and optionally output split reads and/or discordant pairs.\n" "Input sam file must contain sequence header and be grouped by read ids (QNAME).\n" "Input typicallly contains paired-end data, although singleton data is allowed with --ignoreUnmated option.\n" "Output will be all alignments in the same order as input, with duplicates marked with FLAG 0x400.\n\n" "Usage:\n" "For use as a post process on an aligner (eg. bwa mem):\n" " bwa mem samp.r1.fq samp.r2.fq | samblaster [-e] [-d samp.disc.sam] [-s samp.split.sam] | samtools view -Sb - > samp.out.bam\n" " bwa mem -M samp.r1.fq samp.r2.fq | samblaster -M [-e] [-d samp.disc.sam] [-s samp.split.sam] | samtools view -Sb - > samp.out.bam\n" "For use with a pre-existing bam file to pull split, discordant and/or unmapped reads without marking duplicates:\n" " samtools view -h samp.bam | samblaster -a [-e] [-d samp.disc.sam] [-s samp.split.sam] [-u samp.umc.fasta] -o /dev/null\n" "For use with a bam file of singleton long reads to pull split and/or unmapped reads with/without marking duplicates:\n" " samtools view -h samp.bam | samblaster --ignoreUnmated [-e] --maxReadLength 100000 [-s samp.split.sam] [-u samp.umc.fasta] | samtools view -Sb - > samp.out.bam\n" " samtools view -h samp.bam | samblaster --ignoreUnmated -a [-e] [-s samp.split.sam] [-u samp.umc.fasta] -o /dev/null\n" "Input/Output Options:\n" "-i --input FILE Input sam file [stdin].\n" "-o --output FILE Output sam file for all input alignments [stdout].\n" "-d --discordantFile FILE Output discordant read pairs to this file. [no discordant file output]\n" "-s --splitterFile FILE Output split reads to this file abiding by paramaters below. [no splitter file output]\n" "-u --unmappedFile FILE Output unmapped/clipped reads as FASTQ to this file abiding by parameters below. [no unmapped file output].\n" " Requires soft clipping in input file. Will output FASTQ if QUAL information available, otherwise FASTA.\n\n" "Other Options:\n" "-a --acceptDupMarks Accept duplicate marks already in input file instead of looking for duplicates in the input.\n" "-e --excludeDups Exclude reads marked as duplicates from discordant, splitter, and/or unmapped file.\n" "-r --removeDups Remove duplicates reads from all output files. (Implies --excludeDups).\n" " --addMateTags Add MC and MQ tags to all output paired-end SAM lines.\n" " --ignoreUnmated Suppress abort on unmated alignments. Use only when sure input is read-id grouped,\n" " and either paired-end alignments have been filtered or the input file contains singleton reads.\n" "-M Run in compatibility mode; both 0x100 and 0x800 are considered chimeric. Similar to BWA MEM -M option.\n" " --maxReadLength INT Maximum allowed length of the SEQ/QUAL string in the input file. [500]\n" " Primarily useful for marking duplicates in files containing singleton long reads.\n" " --maxSplitCount INT Maximum number of split alignments for a read to be included in splitter file. [2]\n" " --maxUnmappedBases INT Maximum number of un-aligned bases between two alignments to be included in splitter file. [50]\n" " --minIndelSize INT Minimum structural variant feature size for split alignments to be included in splitter file. [50]\n" " --minNonOverlap INT Minimum non-overlaping base pairs between two alignments for a read to be included in splitter file. [20]\n" " --minClipSize INT Minumum number of bases a mapped read must be clipped to be included in unmapped file. [20]\n" "-q --quiet Output fewer statistics.\n"; printVersionString(); fprintf(stderr, "%s", useString); } void printUsageStringAbort() { printUsageString(); exit(1); } // Output the runtine stats. Used both for normal and premature exit. void printRunStats(state_t * state) { // This check should be redundant. if (idCount == 0) { fprintf(stderr, "samblaster: No reads processed.\n"); return; } // Output stats on auxiliary files. if (!state->quiet) { if (state->discordantFile != NULL) fprintf(stderr, "samblaster: Output %10" PRIu64 " discordant read pairs to %s\n", discCount/2, state->discordantFileName); if (state->splitterFile != NULL) fprintf(stderr, "samblaster: Output %10" PRIu64 " split reads to %s\n", splitCount, state->splitterFileName); if (state->unmappedClippedFile != NULL) fprintf(stderr, "samblaster: Output %10" PRIu64 " unmapped/clipped reads to %s\n", unmapClipCount, state->unmappedClippedFileName); } // Now the stats that might indicate a problem that therefore deserve reporting even if quiet set. if (readTooLongCount > 0) { fprintf(stderr, "samblaster: Found %10" PRIu64 " of %10" PRIu64 " (%5.3f) total read ids longer than the --maxReadLength(%d)\n." "samblaster: The longest of which is %d bases long.\n", readTooLongCount, idCount, ((double)100)*readTooLongCount/idCount, state->maxReadLength, readTooLongMax); fprintf(stderr, "samblaster: Consider rerunning samblaster with a larger --maxReadLength.\n"); } if (state->ignoreUnmated) { fprintf(stderr, "samblaster: Found %10" PRIu64 " of %10" PRIu64 " (%5.3f%%) total read ids are marked paired yet are unmated.\n", unmatedCount, idCount, ((double)100)*unmatedCount/idCount); if (unmatedCount > 0) fprintf(stderr, "samblaster: Please double check that input file is read-id (QNAME) grouped.\n"); } if (noPrimaryIdCount > 0) { fprintf(stderr, "samblaster: Found %10" PRIu64 " of %10" PRIu64 " (%5.3f%%) total read ids with no primary alignment.\n", noPrimaryIdCount, idCount, ((double)100)*noPrimaryIdCount/idCount); if (unmatedCount > 0) fprintf(stderr, "samblaster: Please double check that input file is read-id (QNAME) grouped.\n"); } // Now the main output stats. // We need no unmated reads in order to ensure that the various id types form a partition of all read ids. // This is because unmated reads could indicate an incorrectly sorted file which could cause double counting of ids. // Therefore, to be conservative, we should only output the table with both-unmapped, orphan and both-mapped stats in that case. // NOTE: the correct test is the unmated reads instead of --ignoreUnmated, as the latter doesn't mean we actually FIND any unmated. // However, I have decided to output them even if there are unmated reads because we do warn above in that case. // We need to have at least two id types to make sense to output the table. int idCountTypes = 0; if (bothUnmappedIdCount > 0) idCountTypes += 1; if (unmappedOrphanIdCount > 0) idCountTypes += 1; if (mappedOrphanIdCount > 0) idCountTypes += 1; if (bothMappedIdCount > 0) idCountTypes += 1; // We need dups for the denominators to avoid div by zero. if (!state->quiet && dupCount > 0 && idCountTypes > 1) { // We vary the names and spacing of the type names depending on whether or not there are any unmapped orphans. // Use shorter strings with less blank space if there are none, as should be common for properly grouped paired-end data. char * orphanString = (char *)((unmappedOrphanIdCount > 0) ? "Mapped Orphan/Singleton " : "Orphan/Singleton"); int typeLength = strlen(orphanString); fprintf(stderr, "samblaster:\n"); fprintf(stderr, "samblaster: %*.*s Type_ID_Count %%Type/All_IDs Dup_ID_Count %%Dups/Type_ID_Count %%Dups/All_Dups %%Dups/All_IDs\n", typeLength, typeLength, "Pair Type "); fprintf(stderr, "samblaster: %*.*s-----------------------------------------------------------------------------------------------\n", typeLength, typeLength, "------------------------------------------------"); // These first two can't produce any dups. if (bothUnmappedIdCount > 0) { fprintf(stderr, "samblaster: %*.*s %10" PRIu64 " %7.3f %10" PRIu64 " %7.3f %7.3f %7.3f\n", typeLength, typeLength, "Both Unmapped ", bothUnmappedIdCount, ((double)100)*bothUnmappedIdCount/idCount, (UINT64)0, (double)0, (double)0, (double)0); } if (unmappedOrphanIdCount > 0) { fprintf(stderr, "samblaster: %*.*s %10" PRIu64 " %7.3f %10" PRIu64 " %7.3f %7.3f %7.3f\n", typeLength, typeLength, "Unmapped Orphan/Singleton ", unmappedOrphanIdCount, ((double)100)*unmappedOrphanIdCount/idCount, (UINT64)0, (double)0, (double)0, (double)0); } // All orphanDups are, of course, from mapped orphans. if (mappedOrphanIdCount > 0) { fprintf(stderr, "samblaster: %*.*s %10" PRIu64 " %7.3f %10" PRIu64 " %7.3f %7.3f %7.3f\n", typeLength, typeLength, orphanString, mappedOrphanIdCount, ((double)100)*mappedOrphanIdCount/idCount, orphanDupCount, ((double)100)*orphanDupCount/mappedOrphanIdCount, ((double)100)*orphanDupCount/dupCount, ((double)100)*orphanDupCount/idCount); } if (bothMappedIdCount > 0) { fprintf(stderr, "samblaster: %*.*s %10" PRIu64 " %7.3f %10" PRIu64 " %7.3f %7.3f %7.3f\n", typeLength, typeLength, "Both Mapped ", bothMappedIdCount, ((double)100)*bothMappedIdCount/idCount, bothMappedDupCount, ((double)100)*bothMappedDupCount/bothMappedIdCount, ((double)100)*bothMappedDupCount/dupCount, ((double)100)*bothMappedDupCount/idCount); } fprintf(stderr, "samblaster: %*.*s %10" PRIu64 " %7.3f %10" PRIu64 " %7.3f %7.3f %7.3f\n", typeLength, typeLength, "Total ", idCount, ((double)100)*idCount/idCount, dupCount, ((double)100)*dupCount/idCount, (double)100, ((double)100)*dupCount/idCount); fprintf(stderr, "samblaster:\n"); } // Output the main conclusion even if we output the table. // It includes the fact of whether or not the dups were marked or removed, and the timing info. fprintf(stderr, "samblaster: %s %10" PRIu64 " of %10" PRIu64 " (%5.3f%%) total read ids as duplicates", (state->removeDups ? "Removed" : "Marked "), dupCount, idCount, ((double)100)*dupCount/idCount); if ((TIMING == 0) || state->quiet) { fprintf(stderr, ".\n"); } else { #if TIMING struct rusage usagebuf; getrusage(RUSAGE_SELF, &usagebuf); time_t endTime = time(NULL); struct timeval endRUTime = usagebuf.ru_utime; fprintf(stderr, " using %luk memory in ", usagebuf.ru_maxrss); fprintTimeMicroSeconds(stderr, diffTVs(&startRUTime, &endRUTime), 3); fprintf(stderr, " CPU seconds and "); fprintTimeSeconds(stderr, (endTime-startTime), 0); fprintf(stderr, " wall time.\n"); #endif } } /////////////////////////////////////////////////////////////////////////////// // Error Handling. /////////////////////////////////////////////////////////////////////////////// void fatalError(const char * errorStr) { if (idCount == 0) { fprintf(stderr, "%s", errorStr); } else { fprintf(stderr, "%s", errorStr); fprintf(stderr, "samblaster: Exiting early, the following stats are for processing preceeding the error\n"); printRunStats(state); } fprintf(stderr,"samblaster: Premature exit (return code 1).\n"); exit(1); } void fsError(const char * filename) { char * temp; if (errno == ENOENT) asprintf(&temp, "samblaster: File '%s' does not exist.\n", filename); else asprintf(&temp, "samblaster: File system error on %s: %d.\n", filename, errno); fatalError(temp); } /////////////////////////////////////////////////////////////////////////////// // Main Routine. /////////////////////////////////////////////////////////////////////////////// int getIntVal (int argi, int argc, char *argv[]) { char * temp = NULL; if (argi == argc) { asprintf(&temp, "samblaster: Missing value for option %s\n", argv[argi-1]); fatalError(temp); } int retval = str2int(argv[argi]); if (retval == 0 && !streq(argv[argi], "0")) { asprintf(&temp, "samblaster: Invalid integer value '%s' for option %s\n", argv[argi], argv[argi-1]); fatalError(temp); } return retval; } char * getStrVal (int argi, int argc, char *argv[]) { char * temp = NULL; if (argi == argc) { asprintf(&temp, "samblaster: Missing value for option %s\n", argv[argi-1]); fatalError(temp); } return argv[argi]; } int main (int argc, char *argv[]) { // Set up timing. #if TIMING startTime = time(NULL); struct rusage usagebuf; getrusage(RUSAGE_SELF, &usagebuf); startRUTime = usagebuf.ru_utime; #endif // Parse input parameters. state = makeState(); for (int argi = 1; argi < argc; argi++) { // First the general mode and I/O parameters. if (streq(argv[argi], "-h") || streq(argv[argi], "--help")) { printUsageString(); return 0; } else if (streq(argv[argi], "--version")) { printVersionString(); return 0; } else if (streq(argv[argi], "-a") || streq(argv[argi],"--acceptDupMarks")) { state->acceptDups = true; } else if (streq(argv[argi], "-d") || streq(argv[argi],"--discordantFile")) { state->discordantFileName = getStrVal(++argi, argc, argv); } else if (streq(argv[argi], "-e") || streq(argv[argi],"--excludeDups")) { state->excludeDups = true; } else if (streq(argv[argi], "-i") || streq(argv[argi],"--input")) { state->inputFileName = getStrVal(++argi, argc, argv); if (streq(state->inputFileName, "-")) state->inputFileName = (char *)"stdin"; } else if (streq(argv[argi],"--addMateTags")) { state->addMateTags = true; } else if (streq(argv[argi],"--ignoreUnmated")) { state->ignoreUnmated = true; } else if (streq(argv[argi],"-M")) { state->compatMode = true; // In compatibility mode, both 0x100 and 0x800 are considered chimeric and can be used as splitters. // None will be secondary. complementaryBits = (0x100 | 0x800); secondaryBits = 0; } else if (streq(argv[argi],"--maxReadLength")) { state->maxReadLength = getIntVal(++argi, argc, argv); } else if (streq(argv[argi],"--maxSplitCount")) { state->maxSplitCount = getIntVal(++argi, argc, argv); } else if (streq(argv[argi],"--maxUnmappedBases")) { state->maxUnmappedBases = getIntVal(++argi, argc, argv); } else if (streq(argv[argi],"--minClipSize")) { state->minClip = getIntVal(++argi, argc, argv); } else if (streq(argv[argi],"--minIndelSize")) { state->minIndelSize = getIntVal(++argi, argc, argv); } else if (streq(argv[argi],"--minNonOverlap")) { state->minNonOverlap = getIntVal(++argi, argc, argv); } else if (streq(argv[argi], "-o") || streq(argv[argi],"--output")) { state->outputFileName = getStrVal(++argi, argc, argv); } else if (streq(argv[argi], "-r") || streq(argv[argi],"--removeDups")) { state->removeDups = true; } else if (streq(argv[argi], "-q") || streq(argv[argi], "--quiet")) { state->quiet = true; } else if (streq(argv[argi], "-s") || streq(argv[argi],"--splitterFile")) { state->splitterFileName = getStrVal(++argi, argc, argv); } else if (streq(argv[argi], "-u") || streq(argv[argi],"--unmappedFile")) { state->unmappedClippedFileName = getStrVal(++argi, argc, argv); } else { fprintf(stderr, "samblaster: Unrecognized option: %s\n", argv[argi]); printUsageString(); return 1; } } // Make removeDups imply excludeDups if (state->removeDups) state->excludeDups = true; // Output the version number. printVersionString(); // Error check and open files. // Start with the input. if (!streq(state->splitterFileName, "") && state->minNonOverlap < 1) { char * temp; asprintf(&temp, "samblaster: Invalid minimum non overlap parameter given: %d\n", state->minNonOverlap); fatalError(temp); } if (!streq(state->splitterFileName, "") && state->maxSplitCount < 2) { char * temp; asprintf(&temp, "samblaster: Invalid maximum split count parameter given: %d\n", state->maxSplitCount); fatalError(temp); } if (streq(state->inputFileName, "stdin")) { fprintf(stderr, "samblaster: Inputting from stdin\n"); } else { if (!state->quiet) fprintf(stderr, "samblaster: Opening %s for read.\n", state->inputFileName); state->inputFile = fopen(state->inputFileName, "r"); if (state->inputFile == NULL) fsError(state->inputFileName); } if (streq(state->outputFileName, "stdout")) { if (!state->quiet) fprintf(stderr, "samblaster: Outputting to stdout\n"); } else { if (!state->quiet) fprintf(stderr, "samblaster: Opening %s for write.\n", state->outputFileName); state->outputFile = fopen(state->outputFileName, "w"); if (state->outputFile == NULL) fsError(state->outputFileName); } if (!streq(state->discordantFileName, "")) { if (!state->quiet) fprintf(stderr, "samblaster: Opening %s for write.\n", state->discordantFileName); state->discordantFile = fopen(state->discordantFileName, "w"); if (state->discordantFile == NULL) fsError(state->discordantFileName); } if (!streq(state->splitterFileName, "")) { if (!state->quiet) fprintf(stderr, "samblaster: Opening %s for write.\n", state->splitterFileName); state->splitterFile = fopen(state->splitterFileName, "w"); if (state->splitterFile == NULL) fsError(state->splitterFileName); } if (!streq(state->unmappedClippedFileName, "")) { if (!state->quiet) fprintf(stderr, "samblaster: Opening %s for write.\n", state->unmappedClippedFileName); state->unmappedClippedFile = fopen(state->unmappedClippedFileName, "w"); if (state->unmappedClippedFile == NULL) fsError(state->unmappedClippedFileName); } // Read in the SAM header and create the seqs structure. state->seqLens = (UINT32*)calloc(1, sizeof(UINT32)); //initialize to 0 state->seqOffs = (UINT64*)calloc(1, sizeof(UINT64)); //initialize to 0 state->seqs[strdup("*")] = 0; state->seqLens[0] = padLength(0, state); state->seqOffs[0] = 0; int count = 1; UINT64 totalLen = 0; splitLine_t * line; // Read the first line to prime the loop, and also to allow checking for malformed input. line = readLine(state->inputFile); checkBAMfile(line); while (line != NULL && line->fields[0][0] == '@') { // Process input line to see if it defines a sequence. if (streq(line->fields[0], "@SQ")) { char * seqID = NULL; int seqNum = 0; UINT32 seqLen = 0; UINT64 seqOff = 0; for (int i=1; inumFields; ++i) { if (strncmp(line->fields[i], "SN:", 3) == 0) { seqID = line->fields[i]+3; seqNum = count; count += 1; } else if (strncmp(line->fields[i], "LN:", 3) == 0) { seqLen = (UINT32)padLength(str2int(line->fields[i]+3), state); seqOff = totalLen; totalLen += (UINT64)(seqLen+1); } } // Unless we are marking dups, we don't need to use sequence numbers. if (!state->acceptDups) { // grow seqLens and seqOffs arrays if(seqNum % 32768 == 1) { state->seqLens = (UINT32*)realloc(state->seqLens, (seqNum+32768)*sizeof(UINT32)); state->seqOffs = (UINT64*)realloc(state->seqOffs, (seqNum+32768)*sizeof(UINT64)); } state->seqs[strdup(seqID)] = seqNum; state->seqLens[seqNum] = seqLen; state->seqOffs[seqNum] = seqOff; } } // Output the header line. writeLine(line, state->outputFile); if (state->discordantFile != NULL) writeLine(line, state->discordantFile); if (state->splitterFile != NULL) writeLine(line, state->splitterFile); disposeSplitLines(line); // Read in next line. line = readLine(state->inputFile); } // Output the @PG header lines. printPGsamLine(state->outputFile, state); printPGsamLine(state->discordantFile, state); printPGsamLine(state->splitterFile, state); // Make sure we have a header. if (count == 1 && !state->acceptDups) { fatalError("samblaster: Missing header on input sam file. Exiting.\n"); } // Don't count the "*" entry. fprintf(stderr, "samblaster: Loaded %d header sequence entries.\n", count-1); // Make sure we have any more lines to process. if (line == NULL) { fprintf(stderr, "samblaster: No reads in input SAM file.\n"); return 0; } // We can now calculate the size of the signatures array, and intialize it. if (!state->acceptDups) { // Make sure we can handle the number of sequences. int binCount = (totalLen >> BIN_SHIFT); if (binCount >= (1 << 15)) { fatalError("samblaster: Too many sequences in header of input sam file. Exiting.\n"); } state->binCount = binCount; state->sigArraySize = (binCount * 2 + 1) * (binCount * 2 + 1)+1; state->sigs = (sigSet_t *) malloc(state->sigArraySize * sizeof(sigSet_t)); if (state->sigs == NULL) fatalError("samblaster: Unable to allocate signature set array."); for (UINT32 i=0; isigArraySize; i++) hashTableInit(&(state->sigs[i])); } // Now start processing the alignment records. // line already has the first such line in it unless the file only has a header. // Keep a ptr to the end of the current list. splitLine_t * last = line; count = 1; while (true) { splitLine_t * nextLine = readLine(state->inputFile); if (nextLine == NULL) break; if (strcmp(line->fields[QNAME], nextLine->fields[QNAME]) != 0) { processSAMBlock(line, state); last = line = nextLine; count = 1; } else { // If we want the SAM lines to be in the output in the same order as the input.... // we need to be a little careful about how we make the list. last->next = nextLine; last = nextLine; count += 1; } } // We need to process the final set. processSAMBlock(line, state); // Print the runtime stats. printRunStats(state); // Close files. if (!streq(state->inputFileName, "stdin")) fclose(state->inputFile); if (!streq(state->outputFileName, "stdout")) fclose(state->outputFile); if (state->discordantFile != NULL) fclose(state->discordantFile); if (state->splitterFile != NULL) fclose(state->splitterFile); if (state->unmappedClippedFile != NULL) fclose(state->unmappedClippedFile); // Clean up the heap. cleanUpSplitLines(); deleteState(state); freeHashTableNodes(); return 0; } samblaster-v.0.1.26/sbhash.cpp000066400000000000000000000235421366622051700161730ustar00rootroot00000000000000/* -*- Mode: C++ ; indent-tabs-mode: nil ; c-file-style: "stroustrup" -*- Project: samblaster Fast mark duplicates in read-ID grouped SAM file. Also, optionally pull discordants, splitters, and/or unmappend/clipped reads. Author: Greg Faust (gf4ea@virginia.edu) Date: October 2013 File: sbhash.cpp code file for our hash table. License Information: Copyright 2013-2020 Gregory G. Faust Licensed under the MIT license (the "License"); You may not use this file except in compliance with the License. You may obtain a copy of the License at http://opensource.org/licenses/MIT */ #include #include #include #include #include #include #include #include "sbhash.h" /////////////////////////////////////////////////////////////////////////////// // Slab Allocator common code /////////////////////////////////////////////////////////////////////////////// // We will lazily allocate more slabs when needed, and only clean up at the end. ////////////////////////////////////////////////////////////////////////////// void fatalError(const char * errorStr); void checkFSerrWithFilename (ssize_t returnCode) { if (returnCode == -1) { char * temp; if (errno == ENOMEM) temp = (char *)"samblaster: Insufficient memory available to satisfy allocation request.\n"; else asprintf(&temp, "File system error %d trying to allocate or free memory\n", errno); fatalError(temp); } } // Allocate big blocks of memory. char * blockMalloc(ssize_t size) { char * retval = (char *)mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, 0, 0); checkFSerrWithFilename((ssize_t)retval); return retval; } // Free big blocks of memory. Not the size is needed. void blockFree(char * ptr, ssize_t size) { int err = munmap(ptr, size); checkFSerrWithFilename(err); } typedef struct LBMallocBlock LBMallocBlock_t; struct LBMallocBlock { char * block; // Pointer to the payload block LBMallocBlock_t * next; // Pointer to the next allocation block size_t size; // Size of the allocated block }; char * pushNewLBMallocBlock(int blockSize, LBMallocBlock_t **blockArrayPtr) { char * newBlock = blockMalloc(blockSize); LBMallocBlock_t * newMallocBlock = (LBMallocBlock_t *)malloc(sizeof(LBMallocBlock_t)); if (newMallocBlock == NULL) fatalError("samblaster: Insufficeint memory available to allocate (more) objects.\n"); newMallocBlock->size = blockSize; newMallocBlock->block = newBlock; newMallocBlock->next = *blockArrayPtr; *blockArrayPtr = newMallocBlock; return newBlock; } void freeLBMallocBlocks(LBMallocBlock_t * block) { while (block != NULL) { LBMallocBlock_t * nextBlock = block->next; blockFree(block->block, block->size); free(block); block = nextBlock; } } /////////////////////////////////////////////////////////////////////////////// // Hash Table Collision Nodes /////////////////////////////////////////////////////////////////////////////// #define newNodeCount 4096 // Ptr to head of linked list of allocated node slabs. LBMallocBlock_t * nodeBlockList = NULL; // Ptr to head of linked list of free node objects. hashNode_t * hashNodeFreeList = NULL; void makeMoreHashNodes() { hashNode_t * nodeArray = (hashNode_t *)pushNewLBMallocBlock(sizeof(hashNode_t) * newNodeCount, &nodeBlockList); for (int i=1; inext = (nodeArray + i); } (nodeArray + (newNodeCount - 1))->next = NULL; hashNodeFreeList = nodeArray; } hashNode_t * getHashNode() { if (hashNodeFreeList == NULL) makeMoreHashNodes(); hashNode_t * node = hashNodeFreeList; hashNodeFreeList = hashNodeFreeList->next; node->next = NULL; for (int i=0; ivalues[i] = 0; return node; } // I don't think this is currently being called, as we always put the entire string of nodes on the freelist. void disposeHashNode(hashNode_t * node) { node->next = hashNodeFreeList; hashNodeFreeList = node; } void freeHashTableNodes() { freeLBMallocBlocks(nodeBlockList); } /////////////////////////////////////////////////////////////////////////////// // Hash Table /////////////////////////////////////////////////////////////////////////////// // We are going to depend on an old hack. // ptrs to 8 byte things will be 8 byte aligned. // Therefore, the lower 3 bits will be zero. // Also, no known chromosome offset requires all 32 bits. // So we will roll the signature up one bit and put a one there. // We can then tell apart the three following state for a table entry: // 0 -> empty bucket // low bit 1 -> value // low bit 0 -> ptr to overflow nodes. /////////////////////////////////////////////////////////////////////////////// inline UINT64 makeValue(UINT64 value) { return (value << 1) | 1; } inline UINT64 unmakeValue(UINT64 value) { return (value >> 1); } inline hashNode_t * makePtr(UINT64 value) { return (hashNode_t *)value; } inline bool isEmpty(UINT64 value) { return (value == 0); } inline bool isValue(UINT64 value) { return ((value & 1) != 0); } #define numOfSizes 27 static UINT32 hashTableSizes [] = {0, 23, 47, 97, 199, 409, 823, 1741, 3739, 7517, 15173, 30727, 62233, 126271, 256279, 520241, 1056323, 2144977, 4355707, 8844859, 17961079, 36473443, 74066549, 150406843, 305431229, 620239453, 1259520799}; inline UINT32 hash(UINT64 value) { return (UINT32)value; } void hashTableInit(hashTable_t * ht, int size) { ht->entries = 0; ht->size = size; if (size == 0) { ht->table = (UINT64 *)NULL; return; } ht->table = (UINT64 *)calloc(ht->size, sizeof(UINT64)); if (ht->table == NULL) fatalError("samblaster: unable to allocate hash table.\n"); } hashTable_t * makeHashTable() { hashTable_t * ht = (hashTable_t *)malloc(sizeof(hashTable_t)); if (ht == NULL) fatalError("samblaster: unable to allocate hash table.\n"); hashTableInit(ht, hashTableSizes[0]); return ht; } // Use a C++ style destructor so that arrays of hash tables will be cleaned up automagically. hashTable::~hashTable() { if (table != NULL) free(table); } // C style delete. void deleteHashTable(hashTable_t * ht) { if (ht->table != NULL) free(ht->table); } void resizeHashTable(hashTable_t * ht) { // Find out what size table is next. int newsize = 0; for (int i=0; isize) { newsize = hashTableSizes[i+1]; break; } } // Remember the current values array. UINT64 * oldtable = ht->table; int size = ht->size; // Now reinit the hash table with a new table, etc. hashTableInit(ht, newsize); // Now iterate over all values and rehash them into the new table. for (int i=0; ivalues[j]; if (isEmpty(value)) break; hashTableInsert(ht, unmakeValue(value)); } if (node->next == NULL) break; node = node->next; } // We need to free up the nodes. // TODO move out of line. node->next = hashNodeFreeList; hashNodeFreeList = makePtr(oldtable[i]); } // Free up the oldtable. if (oldtable != NULL) free(oldtable); } bool hashTableInsert(hashTable_t * ht, UINT64 value) { // See if we have reached our size limit. if (ht->entries == ht->size) resizeHashTable(ht); int bucket = hash(value) % ht->size; // We need to empty the low order bit so that we can tell the difference between values and ptrs. value = makeValue(value); UINT64 curvalue = ht->table[bucket]; // The empty case should be most common. if (isEmpty(curvalue)) { ht->table[bucket] = value; ht->entries += 1; return true; } // The value case should be next most common. if (isValue(curvalue)) { // The value is already here. if (curvalue == value) return false; // We have a collision and need to add an overflow node. hashNode_t * node = getHashNode(); ht->table[bucket] = (UINT64)node; node->values[0] = curvalue; // Note that this test doesn't cost us anything as it happens at compile time. if (HASHNODE_PAYLOAD_SIZE >= 2) { node->values[1] = value; } else { // We need to add a second new node. hashNode_t * secondNode = getHashNode(); node->next = secondNode; secondNode->values[0] = value; } ht->entries += 1; return true; } // The overflow node case. hashNode_t * curNode = makePtr(curvalue); while (true) { for (int i=0; ivalues[i] == 0) { curNode->values[i] = value; ht->entries += 1; return true; } // Check if the value matches the current value. if (curNode->values[i] == value) return false; } if (curNode->next == NULL) break; curNode = curNode->next; } // If we are here, we need a new node. hashNode_t * node = getHashNode(); curNode->next = node; node->values[0] = value; ht->entries += 1; return true; } samblaster-v.0.1.26/sbhash.h000066400000000000000000000034561366622051700156420ustar00rootroot00000000000000/* -*- Mode: C++ ; indent-tabs-mode: nil ; c-file-style: "stroustrup" -*- Project: samblaster Fast mark duplicates in read-ID grouped SAM file. Also, optionally pull discordants, splitters, and/or unmappend/clipped reads. Author: Greg Faust (gf4ea@virginia.edu) Date: October 2013 File: sbhash.h header file for our hash table. License Information: Copyright 2013-2016 Gregory G. Faust Licensed under the MIT license (the "License"); You may not use this file except in compliance with the License. You may obtain a copy of the License at http://opensource.org/licenses/MIT */ #include #include #include // Rename common integer types. // I like having these shorter name. typedef uint64_t UINT64; typedef uint32_t UINT32; /////////////////////////////////////////////////////////////////////////////// // Hash Table Collision Nodes /////////////////////////////////////////////////////////////////////////////// // This controls the number of data values to store in each hadnNode. #define HASHNODE_PAYLOAD_SIZE 3 typedef struct hashNode hashNode_t; struct hashNode { hashNode_t * next; UINT64 values[HASHNODE_PAYLOAD_SIZE]; }; hashNode_t * getHashNode(); void disposeHashNode(hashNode_t * node); /////////////////////////////////////////////////////////////////////////////// // Hash Table /////////////////////////////////////////////////////////////////////////////// typedef struct hashTable hashTable_t; struct hashTable { UINT64 * table; UINT32 size; UINT32 entries; ~hashTable(); }; hashTable_t * makeHashTable(); void deleteHashTable(hashTable_t * ht); bool hashTableInsert(hashTable_t * ht, UINT64 value); void hashTableInit(hashTable_t * ht, int size=0); void freeHashTableNodes();