A simple consensus approach improves somatic mutation prediction accuracy

Table 4 Sequencing coverage and read-mapping quality features for true-positive and false-positive predictions

Feature	All TPs^a	All FPs^b	DNV^c	Germline^d
Median read depth
Germline	105	25^h	28^h	12^g
Tumor	94	20^h	21^g	19^g
Mean non-reference allele frequency
Germline	0.08%	1.4%^g	1.27%^g	1.77%^g
Tumor	43.10%	30.9%^h	27.3%^h	39.80%
Mean percentage uniquely mapped reads	97.10%	85.3%^f	83.7%^f	89.30%
Fraction of SNVs with <95% reads mapping uniquely	8.3% (19/229)	52.2%^h (71/136)	57.7%^h (56/97)	38.4%^f (15/39)
Fraction of predicted SNVs with >5% of reads mate-rescued	4.4% (10/229)	39.7%^h (54/136)	46.3%^h (45/97)	23.1%^e (9/39)
Mean percentage of reads mapped to multiple locations	1.90%	4.7%^e	4.40%	5.5%^f

Abbreviations: DNV, Did not validate; FP, false positive; J, JointSNVMix2; JS, JointSNVMix2 + SomaticSniper; M, MuTect, MJ, MuTect + JointSNVMix2; MJS, MuTect + JointSNVMix2; + SomaticSniper; MS, MuTect + SomaticSniper; NRAF, non-reference allele frequency; S, SomaticSniper; SNV, single nucleotide variant.; TP, true positive.
^a'Includes all SNVs that validated in the tumor sample.
^bCombined value for these two categories of FPs.
^c Indicates SNVs that were not confirmed in the tumor sample.
^dIndicates SNVs that were also detected in the matched germline.
Significance was tested using Wilcoxon rank-sum test for continuous variables and Fisher’s exact test for fraction of reads mapping uniquely/by mate-pair rescue/to multiple locations: :^eP<0.05; ^fP<1 × 10^-5; ^gP<1 × 10^-10;,,^hP<1 × 10^-15.

ISSN: 1756-994X