Fig. 6From: Achieving robust somatic mutation detection with deep learning models derived from reference data sets of a cancer samplePerformance analysis of the sample-specific models, and comparisons on different INDEL sizes. a Effect of training on the target sample for nine replicates from different SEQC2 data sets. For each sample, two sample-specific models were used: one was trained on only the target sample, and the other with an additional 10% of the training data from the SEQC-WGS-Spike model. b Average performance of somatic mutation callers for different INDEL sizes on SEQC2 replicates. Negative INDEL sizes reflect deletionsBack to article page