Fig. 1: Overview of approach.
From: Accurate somatic variant detection using weakly supervised deep learning
a Matched tumor/normal genomes were used to generate training data. Training labels were generated using high-confidence calls from 4 variant callers (via SMuRF). b Each genomic position selected for training is encoded as a multi-dimensional matrix of reads and associated features (e.g. base quality and mapping quality) and fed to a CNN for training. Source data are provided as a Source Data file.
