KSRMF: Kernelized similarity based regularized matrix factorization framework for predicting anti-cancer drug responses

Abstract

Human Cancer Cell lines have gained a lot of attention since it helps in studying cancer biology and various treatment options. Recently various large-scale drug screening experiments were performed providing access to genomic and pharmacological data. This data helps in predicting drug responses which eventually contributes to the development of personalized cancer treatment. Heterogeneous nature of cancer raises the serious need for therapeutic agents with an essence of personalized treatment. Thus considering the assumption that similar drugs exhibit similar drug responses, we have developed kernelized similarity based regularization matrix factorization framework for predicting anti-cancer drug responses. Drug-Drug chemical structure similarity and Tissue-Tissue similarity (gene expression) are taken as key descriptors to formulate the objective function. The kernel function is used to map non-linear relationships between drugs and tissues. Our aim is to provide an efficient anti-cancer drug response prediction approach to establish the protocol for personalized treatment and new drugs designing. The proposed framework is validated using publicly available tumor datasets: GDSC and CCLE. Proposed KSRMF is further compared with three states of art algorithms using GDSC and CCLE drug screens. We have also predicted missing drug response values in the dataset using KSRMF. KSRMF outperforms other counterparts even though gene mutation data is not incorporated while designing the approach. An average mean square error of 3.24 and 0.504 is achieved using GDSC and CCLE drug screens respectively. The obtained results show that the proposed framework has quite potential to improve anti-cancer drug response prediction. Our analysis showed how data integration can help in achieving the goal of personalized cancer treatment.