HiDRA

Hierarchical Network for Drug Response Prediction with Attention

Model Architecture

HiDRA starts with a dense two-layer drug encoding network. The gene expression data is then split up into pathways based on KEGG data, and each pathway has its own dense encoding network with gene-level and pathway-level attention modules that incorporate the drug encodings. Finally, a dense two-layer network takes concatenated drug and pathway outputs and generates a response prediction.

Feature Representation

• Cancer features:

  • Gene Expression: converted to z-scores for each cell line, genes not present in the KEGG pathway data are removed

• Drug features:

  • Drug Fingerprints: 512-bit Morgan fingerprints generated from SMILES strings with rdkit

URLs

• Original GitHub

• IMPROVE GitHub

References

1. I. Jin and H. Nam. “HiDRA: Hierarchical Network for Drug Response Prediction with Attention”, J. Chem. Inf. Model., 2021