Figure 1.
Overview of the RADIA work-flow for identifying somatic mutations.
The normal DNA, tumor DNA, and tumor RNA BAMs are processed in parallel and initial low-level variants are identified. The variants are filtered by the DNA Only Method using the pairs of normal and tumor DNA and by the Triple BAM Method using all three datasets. The mutations from the two methods are merged and output in VCF format.
Table 1.
DNA Only Method mpileup Support Filters.
Table 2.
Triple BAM mpileup Support Filters.
Figure 2.
Sensitivity of RADIA on simulation data.
Artificial mutations were spiked into the tumor DNA and RNA BAM files of a breast cancer patient using bamsurgeon. (A) Mutations were spiked into the DNA at variant allele frequencies distributed from 1–50% and into the RNA at a constant 25%. The overall sensitivity of RADIA was 85%. RNA Rescue calls from the Triple BAM method detected the mutations that had a DNA VAF less than 10%. (B) Mutations were spiked into the DNA at 10% or less and into the RNA distributed from 1–50%. Most of the DOM mutations are filtered due to the low DNA allelic frequency. The mutations that have adequate RNA read support are rescued back at these low DNA allelic frequencies.
Table 3.
Validation Criteria for Endometrial Carcinoma Data.
Figure 3.
Precision and sensitivity of RADIA on 177 non-hypermutated endometrial carcinoma samples.
Mutations are considered validated in the Somatic Low, Med, or High groups (blue), not validated in the “Not Validated” (green) and Germline/LOH (red) groups, and Ambiguous (orange) when there was low read depth (<10 reads) or ambiguity in the validation data. (A) An overall precision of 98% was demonstrated. RNA Confirmation mutations with strong DNA and RNA support validated over 99%. RNA Rescue mutations validated at 74%. (B) The union of all mutations submitted by TCGA GDACs that validated as somatic was considered as the truth set. RADIA demonstrated an overall sensitivity rate of 84%. Of the mutations that were missed, 33% occurred at low variant allele frequencies (<8%) and 23% occurred in blacklist regions that were ignored.
Table 4.
RNA Rescue Mutations in Lung Adenocarcinoma not Detected by Other Methods in TCGA.