Large-scale comparison of machine learning methods for profiling prediction of kinase inhibitors

Table 1 Detailed ML and DL modelling methods used in this study

Method	Molecular feature	Hyperparameter optimization	Website
RF^a	RDKitDES or fingerprints (Morgan, MACCS, AtomPairs, FP2, and PharmacoPFP)	Grid search	https://github.com/scikit-learn/scikit-learn
NB^b		Grid search	https://github.com/scikit-learn/scikit-learn
SVM^c		Grid search	https://github.com/scikit-learn/scikit-learn
KNN^d		Grid search	https://github.com/scikit-learn/scikit-learn
XGBoost^e		Grid search	https://github.com/dmlc/xgboost
DNN^f		Grid search	https://deepchem.io/
GCN^g	molecular graphs	Grid search	https://deepchem.io/
GAT^h	molecular graphs	Grid search	https://deepchem.io/
MPNNⁱ	molecular graphs	Grid search	https://deepchem.io/
Attentive FP^j	molecular graphs	Grid search	https://deepchem.io/
Chemprop^k	molecular graphs	Bayesian Optimization	https://github.com/chemprop/chemprop
FP-GNN^l	molecular graphs and fixed molecular fingerprints (MACCS, PubChem, and Pharmacophore ErG fingerprints)	Bayesian optimization	https://github.com/idrugLab/FP-GNN

ISSN: 1758-2946