1. Comparison of structure and ligand-based classification models for hERG liability profiling
- Author
-
Serena, Vittorio, Lunghini Filippo, Pedretti Alessandro, Vistoli Giulio, and Beccari Andrea Rosario
- Subjects
machine learning ,Random Forest ,docking ,cardiotoxicity ,hERG ,predictive toxicology - Abstract
The human ether-à-go-go-related potassium channel (hERG) is a voltage-gated potassium channel involved in the repolarization of the cardiac action potential. The off-target inhibition of hERG is the most frequent cause of drug-induced cardiotoxicity. Therefore, assessing hERG related cardiotoxicity in the early phase of the drug discovery process is crucial to avoid undesired cardiotoxic effects. For this purpose, we developed several machine learning classification models for hERG liability profiling basing on Random Forest algorithm by means of Weka software. The models were trained on a dataset of molecules collected from the public repository ChEMBL (https://www.ebi.ac.uk/chembl/) and the commercial GOSTAR database (https://www.gostardb.com/). The training molecules were encoded by both ligand- and structure-based attributes. The former consist of a set of physicochemical descriptors and fingerprints computed by RDKit node available in KNIME, while the latter comprise different scores obtained by docking and rescoring calculations performed by LiGen and Rescore+ tools, respectively. The following models are made available: hERG_LB, trained on ligand-based descriptors hERG_LiGen_AV, trained on a set of scores computed on the docking poses yielded by LiGen, considering for each score the mean value over all the computed poses. hERG_LiGen_AV-LB, trained on the combination of the descriptors used to build hERG_LB and hERG_LiGen_AV-LB models. The input datasets used for the models training and evaluation are made available too.  
- Published
- 2023
- Full Text
- View/download PDF