Authors

Abstract

Transfer learning has been shown to be effective in many applications in which trainingdata for the target problem are limited but data for a related (source) problem are abundant. Weextend the classic transfer learning framework through ensemble and apply it for buildingprediction models of anti-cancer drug response. Previous transfer learning studies for drugresponse prediction focused on building models to predict the response of tumor cells to a specificdrug treatment. We target the more challenging task of building general prediction models that canmake predictions for both new tumor cells and new drugs. Uniquely, we investigate the power oftransfer learning for three drug response prediction applications including drug repurposing,precision oncology, and new drug development, through different data partition schemes in crossvalidation.We implement the ensemble transfer learning framework using LightGBM and twodeep neural network (DNN) models with different architectures, and test them on benchmark invitro drug screening datasets, taking one dataset as the source domain and another dataset as thetarget domain. The analysis results demonstrate the benefit of applying ensemble transfer learningin all three drug response prediction applications with both LightGBM and DNN models.