Fig. 12
From: Implicit-descriptor ligand-based virtual screening by means of collaborative filtering
Performance of RandomForest models with molecular fingerprint Sizes 512, 1024 and 2048 BITS. Illustration of BEDROC20 and EF1% scores on Random Forest models trained on molecular fingerprints across 3 sizes exhibit similar performance across the 3 models. We randomly chose 50 Targets and trained 3 p̈er Targetm̈odels for each Target. Each model was trained with fingerprint sizes of 512 Bits, 1024 Bits and 2048 Bits. The BEDROC20 and EF1% scores were found to be similar across the 3 models for each Target.With little or no discernible difference in the influence of the sizes of the explicit molecular fingerprints towards the performance of the modes, we chose 512 bits to create additional models to compare performance with the collaborative filtering models for our study