VarBin, a novel method for classifying true and false positive variants in NGS data

摘要: Variant discovery for rare genetic diseases using Illumina genome or exome sequencing involves screening of up to millions variants find only the one few causative variant(s). Sequencing alignment errors create "false positive" variants, which are often retained in variant process. Methods remove false positive retain many variants. This report presents VarBin, a method prioritize based on likelihood prediction. VarBin uses Genome Analysis Toolkit calling software calculate variant-to-wild type genotype ratio at each change and position divided by read depth. The resulting Phred-scaled, likelihood-ratio depth (PLRD) was used segregate into 4 Bins with Bin 1 most likely true positive. PLRD values were calculated proband interest 41 additional HiSeq, whole samples (proband's family unrelated samples). At sites without apparent error, wild type/non-variant calls cluster near -3 typically above 10 PLRD. Sites systematic problems (evident quality scores biases as well displayed iGV viewer) tend have higher more variable values. Depending separation proband's value from background same position, method's classification is assigned (Bin 4). To assess performance, Sanger performed 98 samples. True confirmed 97% 30% 2, 0% 3/Bin 4. These data indicate that correctly classifies majority 3/4 contained "uncertain" 2 both Future work will further differentiate 2.