Modeling drug sensitivity: Variable selection, inference and prediction
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A common problem in systems of medicine is to model the sensitivity of drugs based on the genetic characterizations of individuals. Although several predictive models have been developed for drug sensitivity prediction, a recent community effort organized by the Dialogue on Reverse Engineering Assessment and Methods (DREAM) project and the National Cancer Institute (NCI) that explore multiple different drug sensitivity prediction algorithms applied to a common dataset found a Random Forest (RF) based predictive methodology turned out to be the top performer. Although this has been found, certain aspects of the drug prediction problem have still been unresolved. In targeted drug therapy, it is important to identify the genetic features that explains the drug action. As a result, we will construct an algorithm that can both select these super targets and generate a predictive model from those selected features. Furthermore, previous methods of predicting drug sensitivity based on genetic characteristics have been mainly focused on predicting the AUC of the resulting dose-response curves. We have seen that applications in predicting the dose-response curve are limited. In this respect, we will purpose methods to analyze and predict these dose-response curves using a novel stacking algorithm.