swi9 2017:
Note that swi9 study trains callers.
Note that swi9 introduces a new consensus method.
deepSNV and MuTect with the binomial test as germline filter,
SAMtools with -C 200
SiNVICT with option --qscore-cutoff 60
VarScan2 with the parameter --min-var-freq 0.02
default runs from JointSNVMix2, GATK UG, GATK HP, and somaticSniper
Gor4 2016
Four popular somatic SNV caller, i.e. Varscan, SomaticSniper, Strelka and MuTect2 (Supplementary Table S3) were run on above pre-processed sequencing data and with default parameters recommended by the developers. We set the somatic quality threshold of SomaticSniper to 30 (the author recommended 15–40). Raw call sets generated by Varscan and SomaticSniper were filtered by pipelines proposed by the developers, and those generated by Strelka and MuTect2 were processed with built-in post-calling filters for either tool.
Den9 2016
EBcall: default
mutect: validation_strictness Strict
seurat: phred-scaled, somatic score Q > 15
shimmer: base quality > 20, mapping quality > 10
indelocator: Tumor INDEL fraction > 10%, filter by "indel seen by min 2 reads"
sniper: mapping quality > 10, RD: Tumor min 6, normal min 8, SSC > 40
strelka: default
varscan: strand bias, filter FIsher's p-value < 0.05
virmid: mapping quality > 10
bcb8 2015:
varscan: --min-var-freq 0.01
mutect: -U ALLOW_N_CIGAR_READS, --read_filter "NotPrimaryAlignment", --dbsnp, --cosmic, --minimum_mutaion_cell_fraction .10, --enable_qscore_output
freebayes: --genotype-qualities --sctrict-vcf --ploidy, --gvcf, --gvcf-chunk 50000, --min-repeat-entropy 1, --no-partial-observations,
vardict: -c 1, -S 2, -E 3, -g 4, -Q 10,
bro5 2015:
freebayes -f ref.fa aln.bam
samtools mpileup -Euf ref.fa aln.bam — bcftools view -v
java -jar GenomeAnalysisTK.jar -T UnifiedGenotyper -R ref.fa -I aln.bam -stand_call_conf 30 -stand_emit_conf 10 -glm BOTH
java -jar GenomeAnalysisTK.jar -T HaplotypeCaller –genotyping_mode DISCOVERY -R ref.fa -I aln.bam -stand_call_conf 30 -stand_emit_conf 10
Platypus.py callVariants –filterDuplicates=1 –bamFiles= aln.bam –refFile= ref.fa
qia5 2014:
full command line here https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3986649/#S1
note complains that varscan 2 adjusting min allele fraction loses specificity
mutect: --cosmic, --dbsnp
nativesubtract: -dcov 2500, --genotype_likelihoods_model BOTH -minIndelFrac 0.2 --min_base_quality_score 17 --standard_min_confidence_threshold_for_calling 30.0 --standard_min_confidence_threshold_for_emitting 30.0 --baq CALCULATE_AS_NECESSARY --baqGapOpenPenalty 30.0 --defaultBaseQualities -1 --validation_strictness STRICT --interval_merging ALL
sniper: -q 0 -p -F vcf , filters: ad hoc post-calling filtering step with a somatic score cut-off of 20
strelka: default, filters: skipped depth filtration for exome and amplicon sequencing data as recommended by the Strelka authors. For the amplicon sequencing reads, we set the minimum MAPQ score at 17
varscan: mpileups with -q 1 -Q 13 -A -B
multisnv 2015:
We ran SomaticSniper in the joint mode specifying a minimum mappingquality of 30 for reads and leaving other settings as default. To obtain the high confidence dataset (SomaticSniper HC) we used the publicly available Perl scripts (fpfilter.pl and highconfidence.pl) specifying the minimum acceptable base quality (b) as 15, the minimum acceptable mapping quality (q) as 30 and leaving other settings at their default values.
UnifiedGenotyper was ran at default values, specifying a minimum base quality of 20. Events on germline heterozygous sites were filtered out.
MuTect at default values. The high confidence dataset (MuTect HC) consisted of all sites that were not flagged as REJECT in the output VCF file.
Platypus using minimum mapping base quality of 30, minimum base quality of 20, strand bias threshold of 0.01, filtering duplicates –minReads 1 and leaving other settings at default. As with multiSNV, all germline heterozygous sites were filtered out. To get a high confidence dataset from Platypus we kept only sites that were flagged as ‘alleleBiased’ and ‘PASS’.
HapMuC 2014:
Somatic Sniper (1.0.2): - q 15 -Q 0
VarScan 2 (v2.3.6): --min-var-freq 0.05 --min-coverage 10 --min-coverage-normal 10 --min-coverage-tumor 10 --somatic-p-value 0.5
platypus 2014:
Samtools calls were made using Samtools version 0.1.18 (revision 982:295) with the default options. For filtering we used the recommended protocol available at http://sourceforge.net/apps/‐mediawiki/samtools/‐?title=SAM_protocol.
GATK Haplotype Caller calls were made with GATK 2.5, using best practices as described on the GATK website.
aus4 2013:
qSNP 2013:
GATK and Strelka were run in default mode with no changes to default parameters. qSNP was run in standard mode, requiring a minimum of 3 mutant alleles of the same type to make a variant call prior to applying standard
mutect 2013:
Each method was tested in two configurations, standard (STD) and high confidence (HC), with thresholds chosen to produce similar false positive rates across the methods. For SomaticSniper (v1.0.0), we used the published configurations and for JointSNVMix (v0.7.5) we used a detection threshold of P(Somatic) ≥ 0.95 for STD and P(Somatic) ≥ 0.9998 for HC. For Strelka (v0.4.7) we used the recommended configuration with a quality score ≥ 15 for HC and ≥ 1 for STD.
EBcall 2013:
The default setting was applied for running both Genomon-Fisher and VarScan. For SomaticSniper, the -q 30 -Q 15 option was used.
Shearwater 2013:
We ran Caveman as described against a single unmatched normal sample (Papaemmanuil et al., 2013). Similarly, we ran MuTect (v.1.1.4) with default options
—–cosmic b37_cosmic_v54_120711.vcf and —–dbsnp dbsnp_132_b37.leftAligned.vcf.gz
against the same unmatched normal. The options of deepSNV (v.1.3.3) were
combine.method=‘fisher’ and adjust.method=‘BH’
. After calling variants, we filtered the output by removing variants in Ensembl variation (v70) and removed unknown polymorphisms with forumla.
Shimmer 2013:
using Shimmer, JointSNVMix2 (using the train and classify commands), SomaticSniper (with recommended post-filtering), VarScan2 with the ‘somaticFilter’ option and deepSNV. All predicted sSNVs present in dbSNP build 134 were filtered before evaluation of sensitivity and accuracy.
Strelka 2012:
VarScan's ‘somatic’ module was run on the pileup output using the default settings, except when the method was run on the ‘Low Purity’ sample set, in which case a tumor purity setting of 0.4 was provided to indicate the simulated purity of this sample. To reduce VarScan's output to the gain of allele somatic call type analyzed in this study, we filtered all VarScan calls with support for the variant allele in the normal. As a final filtration step, we applied Strelka's high-depth call filter by removing all calls where the normal sample depth was higher than the filtration threshold defined for each chromosome by Strelka. To produce SomaticSniper results, we installed version 1.0.0 of the program bam-somaticsniper and ran this with default settings except for using a minimum mapping quality of 40 to match the cutoff used in VarScan and Strelka. Following the guidelines outlined in the SomaticSniper study, we replicated the ‘Standard’ filtering procedure except that no filtering was performed for calls matching dbSNP entries. This exception was made because we use dbSNP overlap to evaluate somatic call quality for the various workflows below.
Seurat 2012:
varscan: java -jar ./bin/VarScan.v2.3.4.jar somatic $1.pileup $2.pileup $3 --outputvcf --min-coverage-normal 6 --normal-purity $4 --tumor-purity $5
$4 and $5 are values for expected normal and tumor purity (1.0 being the default for both, meaning zero contamination). Since a) VarScan does not include a method to estimate the tumor purity and b) the other software is agnostic to the true level (and dimensionality) of tumor heterogeneity, these values were left to their defaults.
strelka: strelka_workflow/configureStrelkaWorkflow.pl --normal=$1 --tumor=$2 –ref=human_g1k_v37_short_sequence_name.fasta --config=strelka_config_bwa_default.ini --output-dir=$3
sniper: ./bin/bam-somaticsniper -F vcf -f human_g1k_v37_short_sequence_name.fasta $2 $1 $3 -q 10 -Q 10