A novel network controllability algorithm to target personalized driver genes for discovering combinational drugs of individual cancer patient

Abstract

It is a big challenge to identify patient-specific drug combinations based on cancer omics data. However, most conventional methods used for identifying personalized therapies require large sample sizes and focuse on the population as a whole and we still lack a feasible mathematical framework to circumvent this key problem in the clinical application of precision medicine. This work presents a personalized drug controller method (PDC) to identify drug combinations of individual cancer patients, by exploring the transition state information from a disease state to a normal state, thus providing novel insights into tumor heterogeneity in complex patient ecosystems.We validate the effectness of PDC in terms of the accurate prediction of drug combinations on two benchmark cancer datasets. We can also discover personalized key control genes (KCGs) even if the KCGs are hidden in transcription profiles by exploring their network and structural characteristics. Furthermore, we provide computationally derived side effect signatures for drug combinations based on patient-specific KCGs to enhance patient stratification and prognostication. The experimental results strongly support that PDC is effective for personalized drug therapy and treatment, and can fill the gap between network control theory and the personalized drug discovery problem.