Skip to main content
Fig. 1 | Genome Biology

Fig. 1

From: Achieving robust somatic mutation detection with deep learning models derived from reference data sets of a cancer sample

Fig. 1

Overall performance of different NeuSomatic network trained models on 119 replicates from the SEQC2 data set. a The models trained on SEQC2 data achieved consistent superiority over the DREAM model across diverse sets of replicates of different purities/coverages in WGS, WES, FFPE, AmpliSeq (AMS), and different library preparation data sets. In this subfigure, for each replicate, the best F1-score was computed across different approaches. The heatmaps illustrate the absolute difference between the F1-score of any of the network models according to the best F1-score. In each panel, the mean F1-score is shown for each approach across 119 replicates. b Precision-recall analysis of different network models on different SEQC2 data sets. Each datapoint shows the average precision and recall values for SNVs and INDELs across the samples in a given dataset. c F1-score (%) comparison for different network models across different coverages (10×–300×) and tumor purities (5–100%). For a given coverage, results of different models on the same tumor purity are connected for better illustration.

Back to article page