Abstract
We present a novel approach for the prediction of anticancer compoundsensitivity by means of multi-modal attention-based neural networks (PaccMann).In our approach, we integrate three key pillars of drug sensitivity, namely,the molecular structure of compounds, transcriptomic profiles of cancer cellsas well as prior knowledge about interactions among proteins within cells. Ourmodels ingest a drug-cell pair consisting of SMILES encoding of a compound andthe gene expression profile of a cancer cell and predicts an IC50 sensitivityvalue. Gene expression profiles are encoded using an attention-based encodingmechanism that assigns high weights to the most informative genes. We presentand study three encoders for SMILES string of compounds: 1) bidirectionalrecurrent 2) convolutional 3) attention-based encoders. We compare our devisedmodels against a baseline model that ingests engineered fingerprints torepresent the molecular structure. We demonstrate that using ourattention-based encoders, we can surpass the baseline model. The use ofattention-based encoders enhance interpretability and enable us to identifygenes, bonds and atoms that were used by the network to make a prediction.