idti_RBM.zip

Side effect is one of the major causes of failure in drug development. Severe side effect that go undetected until the post-marketing phase a drug often lead to patient morbidity. Therefore, there is an urgent need for a method to identify potential side effects and predict the side effects of new drug. Following this need, we present a new predictor of drug-side effect associations. First, we construct multiple similarity matrices from drug space and side-effect space, respectively. Second, these similarity matrices are linear weighted by optimized Centered Kernel Alignment-based Multiple Kernel Learning (CKA-MKL) algorithm in two different spaces. Then, Weighted K nearest known neighbors (WKNKN) is utilized to preprocess the association matrix. Next, we construct Restricted Boltzmann machines (RBM) in drug space and side effect space, respectively, and apply a penalized maximum likelihood approach to train model. At last, we adopt the average scoring rule to integrate predictions from RBMs. We demonstrate, with three benchmark datasets, that our method is able to give a more stable and accurate prediction results.