Datasets for training and for testing ligand-based reverse screening to predict drug targets.

Bioactivity data were obtained from the ChEMBL (version 25) and the Reaxys® (version 03.2019) databases for training/screening and testing target prediction of small molecule drug by reverse screening. A short extract of the raw ChEMBL data for training is given in Supplementary Table 1 (Supp_Table_1.xlsx) to show three lines corresponding to an active, an inactive and an ‘gray area’ datapoints, respectively. Processed data are deposited here. The screening set file contains (CHEMBL-screeningset.csv), for each active compound, the standardized SMILES, the ChEMBLID, the number of experimental target(s) and their UniProt identifier(s). The test set file (Reaxys-testset-with300-SMILES.csv) contains, for each 364201 active compound, the ReaxysID, the number of experimental target(s) and the UniProt identifier(s). For Reaxys users, the chemical structure can be obtained through bulk request on the website. Access to www.reaxys.com and to Reaxys data can be obtained by contacting Elsevier directly. The first 300 entries display the standardized SMILES so that every reader can reproduce the results obtained by the reverse screening exercise.

Further datasets can be obtained at www.swisstargetprediction.ch/download.php