PathDSP

Explainable Drug Sensitivity Prediction through Cancer Pathway Enrichment Scores

Model Architecture

PathDSP first converts the cancer and drug features into pathway-level enrichment scores using a network-based approach, then feeds those into a fully connected neural network (FNN) with four hidden layers.

Feature Representation

• Cancer features:

  • Gene Expression: converted into pathway-level activity using ssGSEA algorithm

  • CNV: converted into pathway-level activity using NetPEA algorithm

  • Mutation: converted into pathway-level activity using NetPEA algorithm

• Drug features:

  • SMILES: converted into morgan fingerprints

  • Drug target: converted into pathway-level activity using NetPEA algorithm

URLs

• Original GitHub

• IMPROVE GitHub

References

1. Y. Tang and A. Gottlieb. “Explainable drug sensitivity prediction through cancer pathway enrichment”, Scientific Reports, 2021.