Fig. 5
From: Comprehensive machine learning boosts structure-based virtual screening for PARP1 inhibitors
Screening performance of SFs on the full and dissimilar test set versions in terms of NEF1%. The SFs are: generic SFs (A), classification-based target-specific ML SFs (B), and regression-based target-specific ML SFs (C). The NEF1% of each ML SF-test set pair is the median obtained after 10 training-test runs of the corresponding learning algorithm on the respective test set. Random NEF1% values for the full and dissimilar test set versions are 0.020 and 0.017, respectively