DeepSom is a convolutional neural network (CNN) - based pipeline for calling somatic SNP and INDEL variants without a matched normal.
DeepSom can operate in three modes:
Inference mode. In this mode, the convolutional neural network (CNN) assigns a pseudoprobability score to each candidate variant which can then be classified as somatic or non-somatic. This is the main operating mode.
Test(evaluation) mode. In this mode, we use variants with known labels (somatic/non-somatic) to evaluate DeepSom performance.
Train mode. In this mode, the CNN learns relationships between variant tensors and variant labels (somatic/non-somatic). This results in a model which can be used in inference or evaluation.
See dataprep on how to prepare data for the CNN.
See cnn on how to run the CNN.
See test for a test run of DeepSom on a preprocessed set of variants.