QSAR-derived affinity fingerprints (part 1): fingerprint construction and modeling performance for similarity searching bioactivity classification and scaffold hopping

摘要

The workflow for the calculation of the rv-QAFFP fingerprint. 1360 ligand sets (Additional file ) assayed against various molecular targets were extracted from the ChEMBL19 database [ , ]. For each ligand set, Random Forest model was built using 80% of data for training and 20% for testing. Each QSAR model was validated using both internal (i.e., cross-validated) and external (i.e., test set) error measures and only models that satisfied stringent quality criteria were used for the construction of the rv-QAFFP fingerprint. The applicability domain of individual QSAR models was estimated using inductive conformal prediction [ – ]. The rv-QAFFP fingerprint is composed of 440 affinities predicted for the panel of assays covering 376 distinct molecular targets