Fig. 1

Overall process of RF-QSAR. First of all, 1121 target models are built by bioactivity data from ChEMBL database. As a user input a query ligand to the server, scores for target models are calculated to build a score vector. Then, the score vector is transformed into the probabilities to be active. Finally, top-k targets are proposed ranked by their probabilities to the query ligand. Targets to search can be filtered by their classes according to user’s preference