Interpretable deep learning for cancer personalized medicine

In recent years, deep learning models have resulted in outstanding breakthrough performances. However, many models behave as black boxes that can hide data biases, incorrect hypotheses or even software errors. In this talk, Maria Rodriguez Martinez (IBM Zürich) illustrates how interpretable deep learning models can achieve both high prediction accuracy and transparency.

First, she will introduce multi-modal deep learning models that predict drug response while highlighting the genetic and chemical patterns that were more informative to make a prediction. Rodriguez Martinez also discusses how reinforcement learning approaches can facilitate the early phases of drug discovery and support the personalized design of new candidate compounds.

Focusing next on T cell-based immunotherapies, she presents a model to predict the binding of T cell receptors and epitopes. This model can be coupled with an easy-to-use interpretable pipeline to extract the binding rules governing the T cell binding. These approaches are a first step towards the design and engineering of receptors of improved affinity.

Finally, she will briefly discuss how interpretable models enable the linkage of different data modalities, such as image and molecular data. In doing so, they can pave the way to the development of personalized models of therapeutic interventions.