Fig. 6
From: TB-IECS: an accurate machine learning-based scoring function for virtual screening
The impact of feature vector length and interaction energy terms on model performance. A: the change of the model performance after feature vector length reduction, B: the change of the model performance after feature vector length increasing. The feature reduction was implemented by remove descriptors of NNscore from raw feature combinations while the feature vector length was increased through decomposed the raw energy terms into interaction fingerprints