Sparse discriminative latent characteristics for predicting cancer drug sensitivity from genomic features

Drug screening studies typically involve assaying the sensitivity of a range of cancer cell lines across an array of anti-cancer therapeutics. Alongside these sensitivity measurements high dimensional molecular characterizations of the cell lines are typically available, including gene expression, copy number variation and genomic mutations. We propose a sparse multitask regression model which learns discriminative latent characteristics that predict drug sensitivity and are associated with specific molecular features. We use ideas from Bayesian nonparametrics to automatically infer the appropriate number of these latent characteristics. The resulting analysis couples high predictive performance with interpretability since each latent characteristic involves a typically small set of drugs, cell lines and genomic features. Our model uncovers a number of drug-gene sensitivity associations missed by single gene analyses. We functionally validate one such novel association: that increased expression of the cell-cycle regulator C/EBPδ decreases sensitivity to the histone deacetylase (HDAC) inhibitor panobinostat.Author summary: A core tenant of precision medicine is that treatment should be tailored to the patient. In the context of cancer, large-scale screens, assaying the sensitivity of many cell-lines to panels of drugs, have the potential to enable discovery of biomarkers of sensitivity to specific therapeutics. However, existing computational approaches have not taken full advantage of these data. We develop a novel multi-task regression model, Lacrosse, which uses a Bayesian non-parametric prior to model “latent characteristics” of cell-lines that confer sensitivity to specific drugs and are predictable from genomic features. The resulting algorithm improves upon existing work by: a) jointly modeling multiple drugs to share statistical signal b) incorporating prior knowledge in terms of known inhibition targets c) using a sparse latent variable regression approach giving interpretable summaries of detected gene-drug associations. In particular, our analysis uncovers groups of drugs whose efficacy depends on genomic features in a similar way. We find new potential biomarkers of drug sensitivity, one of which we validate experimentally: that panobinostat is less effective when C/EBPδ is over-expressed.