Skip to main content

Table 5 Additional filtering improved specificity of predictions lacking full consensus

From: A simple consensus approach improves somatic mutation prediction accuracy

  Validation ratea    TP dropout rated
  Partial consensusb No consensusc Overall  
Base validation ratee 44/78 (56%) 6/105 (6%) 50/183 (27%)  
Percentage of mate-rescued reads <7%f 44/64 (64%) 6/69 (9%) 50/133 (38%) 0%
RD >10 (tumor and germline)g 40/68 (59%) 5/72 (7%) 45/140 (32%) 10%
Germline non-reference allele frequency ≤0.02h 42/65 (65%) 6/75 (8%) 48/140 (34%) 4%
GATK prediction for SNV in tumori 43/66 (66%) 5/47 (11%) 48/113 (42%) 4%
GATK + mate-rescued 43/56 (77%) 5/31 (16%) 48/87 (55%) 4%
GATK + mate-rescued + RD >10 39/46 (85%) 4/19 (21%) 43/65 (66%) 14%
  1. Abbreviations: RD, read depth; SNV, single nucleotide variant; TP, true positive.
  2. aValidation rate = TPs/total SNVs assessed.
  3. bPartial-consensus predictions (made by two programs).
  4. cNo consensus predictions (made by only one program).
  5. dPercentage of TPs that would be discarded if the indicated set of filters was applied, that is, loss in sensitivity.
  6. eRefers to specificity prior to filtering.
  7. fFiltering on percentage of reads made from mate-pair rescue.
  8. gFiltering on RD increased from 8 to 10 in tumor and germline.
  9. hNon-reference allele frequencies in germline decreased from ≤0.05 to ≤0.02.
  10. iFiltering SNVs predicted in the tumor but not the germline by GATK Unified Genotyper.