Table 1 Comparison of model performance56,57,58,59,60

From: Resolving data bias improves generalization in binding affinity prediction

  1. Reported CASF2016 scoring performance (Pearson correlation coefficients and r.m.s.e. values) of all published deep-learning-based scoring functions that have, to our knowledge, evaluated their scoring performance on the complete CASF2016 (n = 285) dataset. Models with dark grey bars and bold text have been trained or created in this study (Pafnucy, GenScore and our two search algorithms ‘Search top 5 complexes’ and ‘Search top 5 ligands’). White bars represent published binding affinity prediction models, with performance values taken from literature. Models with light-grey bars are classical scoring functions (AutoDock Vina, GlideScore). As a baseline, the bottom bar in the r.m.s.e. table shows the error that is achieved when the average training dataset label is assigned to all CASF2016 complexes.
  2. Source data
Back to article page