High-throughput screening (HTS) of genomically characterized cancer cell lines is used to profile genetic dependencies or drug responses. We introduced the SuperDendrix algorithm to examine associations between genetic dependencies in cancer cell lines and find dependencies explained by combinations of mutually exclusive somatic mutations congregating into a few oncogenic pathways across cancer subtypes. Superdendrix uses an integer linear program to find (approximately) mutually exclusive genomic alterations that associate with a quantitative phenotype. In earlier related work we developed LOBICO, an algorithm to explain continuous output like drug responses of HTS experiments by small Boolean formulas. Recently we also evaluated machine learning algorithms like multi-output support vector regression or graph neural networks (GNNs) on large-scale drug screens to incorporate the relation between output vectors for different drugs or between the molecular structure of a drug and the effect on a cell line, respectively. Current work includes building a GNN that uses molecular similarities based on functional groups to predict the effect of a new drug on a cell line.
For internal attendees
Symbiose seminars : https://www.cesgo.org/symbiose/seminars/explainable-and-predictive-mode…