training using bamsurgeon on multiple samples

[kiranchari]

Hi,

I generated a training set comprising multiple samples using the method suggested in the README (using somaticseq's docker pipeline for bamsurgeon): https://github.com/bioinform/neusomatic#creating-training-data

After training Neusomatic on this dataset, I do not get the expected performance on an external test sample independent of the training set - I am not sure where the issue is. I have previously evaluated Neusomatic's output without issues.

One issue I suspect is with the vcf generated by bamsurgeon. While I can view the alignments at the mutated sites in the generated insilico normal/tumor samples, bamsurgeon reports a DPR (depth) of 0 for most mutations in the VCF. Below is a sample of the synthetic vcf generated by bamsurgeon. Could this be an issue for Neusomatic? i.e. does it rely on the DPR reported in the VCF when generating training data? The insilico samples look fine otherwise, and the Neusomatic training ran smoothly for 1000 epochs.

Thank you.

##fileformat=VCFv4.1
##phasing=none
##INDIVIDUAL=TRUTH
##SAMPLE=<ID=TRUTH,Description="bamsurgeon spike-in",Individual=TRUTH>
##INFO=<ID=CIPOS,Number=2,Type=Integer,Description="Confidence interval around POS for imprecise variants">
##INFO=<ID=IMPRECISE,Number=0,Type=Flag,Description="Imprecise structural variation">
##INFO=<ID=SVTYPE,Number=1,Type=String,Description="Type of structural variant">
##INFO=<ID=SVLEN,Number=.,Type=Integer,Description="Difference in length between REF and ALT alleles">
##INFO=<ID=SOMATIC,Number=0,Type=Flag,Description="Somatic mutation in primary">
##INFO=<ID=VAF,Number=1,Type=Float,Description="Variant Allele Frequency">
##INFO=<ID=DPR,Number=1,Type=Float,Description="Avg Depth in Region (+/- 1bp)">
##INFO=<ID=MATEID,Number=1,Type=String,Description="Breakend mate">
##ALT=<ID=INV,Description="Inversion">
##ALT=<ID=DUP,Description="Duplication">
##ALT=<ID=DEL,Description="Deletion">
##ALT=<ID=INS,Description="Insertion">
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT SPIKEIN
1 139455 . A G 100 PASS SOMATIC;VAF=0.288888888889;DPR=0.0 GT 0/1
1 148410 . C A 100 PASS SOMATIC;VAF=0.123595505618;DPR=0.0 GT 0/1
1 148416 . T G 100 PASS SOMATIC;VAF=0.239583333333;DPR=0.0 GT 0/1
1 822932 . G C 100 PASS SOMATIC;VAF=0.0903225806452;DPR=0.0 GT 0/1
21 10569712 . AT A 100 PASS SOMATIC GT 0/1
21 10569714 . TGA T 100 PASS SOMATIC GT 0/1
21 10703018 . C CCCT 100 PASS SOMATIC GT 0/1
21 10706729 . G GA 100 PASS SOMATIC GT 0/1