SqCL - target capture pipeline

Scripts to work with conserved loci data from squamates.

Areas of future improvements

• matching of contigs to targets has some weird behavior
  • check cases with high matches where matches aren't retained
• use read error correction prior to assembly
  • http://www.genome.umd.edu/quorum.html
  • check with M. Grundler about this
• use cap3 to make PRG rather than random picking
• better trusted SNP set for BQSR
  • maybe do Cortex intersection with raw GATK call
• check SNP calling filters?
  • https://www.broadinstitute.org/gatk/guide/article?id=3225
  • https://www.broadinstitute.org/gatk/guide/article?id=6925
• check mito DNA first both for misidentification & for contamination
• use kmer alignment-free methods to calculate distances to look for possible contaminants or misidentification (MASH or kWIP)
• add more plotting to summarize SNP quality
  • heterozygosity vs. AF
  • coverage vs. heterozygosity
• add more plotting to identify collapsed paralogs or chimeric sequences
  • sliding window plots of p-distance across loci
  • sliding window plots of heterozygosity across loci
  • sliding window plots of depth across loci
  • from these, identify peaky distributions

Notes before you start

• When running any script, you can see the script arguments using python script.py --help.
• This pipeline is best run with these scripts all in a row, but each script can be run on its own. However, this piecemeal functionality hasn't been tested.

Steps

1. Start with demultiplexed data.

2. Create a CSV file summarizing data

   • note, column order does not matter.
   • sample: sample name
   • read1: complete path to read 1 file
   • read2: complete path to read 2 file
   • adaptor1: adaptor 1 to scrub; put asterisk where barcode goes
   • adaptor2: adaptor 2 to scrub; put asterisk where barcode goes
   • barcode1: barcode 1 to scrub
   • barcode2: barcode 2 to scrub (if it exists)
   • lineage: the group to which the sample belongs; used for assembly and SNP calling

3. Clean reads

   • Remove adaptors using Trimmomatic
   • Merge reads using PEAR
   • Lightly remove low quality sequence using Trimmomatic
   • Ends up creating three files: 2 paired files with ordering retained, and 1 unpaired read file
   • Assumes: Trimmomatic v0.36 and PEAR 0.9.10

   python ~/squamateUCE/clean_reads.py --trimjar ~/bin/Trimmomatic-0.36/trimmomatic-0.36.jar \
   	--PEAR ~/bin/pear-0.9.10/pear-0.9.10-bin-64 --dir /scratch/drabosky_flux/sosi/uce_test/ \
   	--sample Anolis_carolinensis --file /scratch/drabosky_flux/sosi/uce_test/samples.csv
   

   IF YOU WANT This is a good place to check data quality by running quality_1_reads.py.

4. Assemble reads

   • Trinity is meant for transcriptomes, but the authors of Phyluce have found it works well with non-transcriptome data sets
   • This script combines read 1 and unpaired reads into one file to use with Trinity
   • Then assembles
   • This is very memory intensive - recommended to give 1 Gb of RAM per 1M reads
   • Can use --normal flag to normalize big data sets
     • found this to be necessary because otherwise had heap error during Butterfly step
   • Requires Trinity v2.2.0

   python /home/sosi/squamateUCE/trinity_assembly.py --trinity ~/bin/trinityrnaseq-2.2.0/Trinity \
   	--sample Anolis_carolinensis --dir /scratch/drabosky_flux/sosi/uce_test/ \
   	--mem 55 --CPU 16
   

5. Match assemblies to original targets

   • Uses blat and a soft reciprocal match to identify which probes match to which contigs in the assembly
   • Multiple probes belong to the same locus, and while the probes don't necessarily overlap, they are likely to match to the same contig in the assembly
   • Script uses a regex that is dataset specific to synonymize probes to targeted loci
   • This script only considers matches that are above the e-value provided by the user and that are no more than 10 orders of magnitude worse than the best match
   • Have a few kinds of outcomes
     • easy_recip_match: 1-to-1 unique match between contig and targeted locus
     • complicated_recip_match: 1-to-1 non-unique match, in which one targeted locus matches to multiple contigs
     • ditched_no_recip_match: a case in which the contig matches to the targeted locus, but it isn't the best match
     • ditched_no_match: a case in which the contig matches to the targeted locus, but the locus doesn't match to the contig
     • ditched_too_many_matches: a case in which one contig has multiple good matches to multiple targeted loci
       • removed because possible paralogs?
   • This requires blat v36

   python ~/squamateUCE/match_contigs_to_probes.py --blat ~/bin/blat --sample Anolis_carolinensis \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --evalue 1e-30 \
   	--db squamate_AHE_UCE_genes_loci.fasta
   

6. Generate pseudo-reference genomes

   • Uses lineage designation in sample file created in step 2 to create a pseudo-reference genome (PRG) for each lineage
   • The script identifies all contig matches to all targeted loci across all individuals in that lineage
     • if there are multiple contigs across individuals that match to a given targeted locus, then:
       • either picks the top contig if it is much better than the other matches (>1e3)
       • or if not, picks the longest one
   • This script can be run with only the 'easy' and / or 'complicated' reciprocal matches
   • Also will reverse complement contigs so that they are ordered the same as the contig (important later for phylogeny making!)
   • Requires no additional programs.

   python ~/squamateUCE/make_PRG.py --lineage l1 --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --keep easy_recip_match
   

   IF YOU WANT This is a good place to check data quality by running quality_2_assembly.py.

7. Align reads

   • align_reads1.py: run by individual
     • Align paired reads & unpaired reads in two separate runs using bwa
       • paired read run performs mate rescue
     • Then run fixmate to fix any broken paired-end information
     • Then sort the BAM files
     • Then mark duplicates
   • align_reads2.py: run by lineage first, then by individual
     • Generates raw set of variants using bcftools for the lineage
     • Filters raw set lightly (--dp and --qual)
     • Uses filtered set to perform base quality score recalibration (BQSR)
     • Recalibrates individual BAM files
   • These scripts asssume BWA 0.7.12, samtools 1.3.1, GATK 4, Picard 2.23, and bcftools 1.7

   python ~/squamateUCE/align_reads1.py --sample Mus_musculus \ 
   	--file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --bwa ~/bin/bwa-0.7.12/bwa \
   	--samtools ~/bin/samtools-1.3.1/samtools \
   	--picard ~/bin/picard-tools-2.23.1/picard.jar --CPU 1 --mem 1
   	
   python ~/squamateUCE/align_reads2.py --lineage l1 --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --samtools ~/bin/samtools-1.3.1/samtools \
   	--bcftools ~/bin/bcftools-1.7/bcftools \
   	--gatk ~/bin/GenomeAnalysisTK.jar --mem 3 --dp 10 --qual 20 --CPU 4
   

   IF YOU WANT This is a good place to check data quality by running quality_3_alignment.py.

8. Call and filter variants

   • Call variant (and invariant sites!) based on "final" BAM files per lineage
     • Note the importance of calling invariant sites - it is the denominator in all pop gen analyses
   • Variants are filtered based on quality (--qual)
   • And then filtered on depth, on a per individual basis (--dp)
     • If an individual has too low of coverage, 'ALLELE/ALLELE' becomes missing ('./.')
     • If every individual in the lineage is missing, then the site is dropped
   • This requires bcftools

   python ~/squamateUCE/call_variants.py --lineage l1 --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --bcftools ~/bin/bcftools-1.7/bcftools --mem 4 \
   	--CPU 4 --dp 10 --qual 20
   

Phylogeny Making

1. Make alignments.
   • Takes in all the data in the sample file to identify all unique pseudo-reference genomes
   • Creates unaligned fasta files for any locus that is sampled for four or more individuals
     • because most gene tree - species tree programs take in unrooted gene trees, there is no information in three-taxon trees
   • Also creates a summary file that has information about missingness

   python ~/squamateUCE/phylogeny_make_alignments.py --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/
   

2. Actually do the alignments and make the gene trees.
   • Does the alignments and makes the gene trees for all files.
   • Uses multiprocessing module in Python to do it more quickly
   • Assumes the user has mafft 7.294 and RAxML 8.2.4

   python ~/squamateUCE/phylogeny_align_genetrees.py --dir /scratch/drabosky_flux/sosi/uce_test/ \
   	--CPU 4 --mafft ~/bin/mafft --raxml ~/bin/standard-RAxML/raxmlHPC
   

3. Make the concatenated alignment.
   • Designed to be used with RAxML, but could work with any phylogeny program.

   python ~/squamateUCE/phylogeny_make_concatenated.py --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
   	--dir /scratch/drabosky_flux/sosi/uce_test/ --miss 0.8
   

4. Makes the files to be used for ASTRID and ASTRAL.
   • Creates a file with all the best trees for each gene.
   • And another file that has the file paths to the bootstrap trees.

    python ~/squamateUCE/phylogeny_prep_astrid_astral.py --file /scratch/drabosky_flux/sosi/uce_test/samples.csv \
    	--dir /scratch/drabosky_flux/sosi/uce_test/ --miss 0.8
   

   • Can then run with commands like these:

   java -jar astral.4.10.6.jar -i best_trees_0.9.trees -b bootstrap_files_0.9.txt -r 100 -o astral_0.9.tre
   ASTRID -b bootstrap_files_0.9.txt -i best_trees_0.9.trees -o astrid_0.9.tre