Proceeding talk – Theme: Data.
Abstract
Drug response prediction is valuable for developing hypotheses for selecting therapies tailored for individual cancer patients. However, the prediction task is extremely challenging, raising the need for methods that can effectively model and predict drug responses in cancer. We propose a novel formulation of multi-task matrix factorization that allows selective data integration for predicting drug responses. Our method combines Bayesian matrix factorization with component-wise multiple kernel learning (MKL). We show an improved predictive performance of the proposed method, on two publicly available cancer data sets and a synthetic data set. In addition, we validate the predictive abilities of the model with independent lab experiment using an in-house cancer cell line panel. Finally, we show that the proposed method exploits the prior knowledge (known pathway information) in a meaningful way to learn the drug response associations, opening up the opportunity for elucidating drug action mechanisms.
Authors
Muhammad Ammad-Ud-Din, Helsinki Institute for Information Technology HIIT, Department of Computer Science, Aalto University, Finland
Suleiman A.Khan, Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland
Disha Malani, Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland
Astrid Murumägi, Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland
Olli Kallioniemi, Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland
Tero Aittokallio, Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland
Samuel Kaski, Helsinki Institute for Information Technology HIIT, Department of Computer Science, Aalto University, Finland