Target Predictions using LINCS Data

Identifying the binding targets of small molecules is an essential process in drug discovery and development. The two conventional approaches include high throughput screening (HTS) and computational structural docking. HTS suffers from its expensive cost and time-consuming procedure, while the computational methods reply on simplifying assumptions that often leads to less accurate results. In this project, we developed machine learning based approaches to efficiently predict drug targets using the massive LINCS data. We extracted meaningful features from the LINCS data and integrated them with information from other genomic data, and build a random forest based classifier that achieves remarkable prediction accuracy. Our strategy provide an fast and efficient way of predicting drug targets, and can naturally serve as a pre-pruning step for the computationally expensive structural based approaches.