Classification of Metal Binders by Naïve Bayes Classifier on the Base of Molecular Fragment Descriptors and Ensemble Modeling. Issue 6 (10th April 2019)

• Record Type: Journal Article
• Authors: Solov'ev, Vitaly; Tsivadze, Aslan; Marcou, Gilles; Varnek, Alexandre
• Is Part Of: Molecular informatics. Volume 38:Issue 6(2019)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2019-04-10
• Subjects: Naïve Bayes classifier -- ensemble classification -- structure-property modeling -- consensus model; molecular fragment descriptors -- complexation of metal ion with organic ligand -- metal binder; stability constant; Cheminformatics -- Periodicals; QSAR (Biochemistry) -- Periodicals; Structure-activity relationships (Biochemistry) -- Periodicals; Drugs -- Structure-activity relationships -- Periodicals; 615.19
• DOI: 10.1002/minf.201900002
• Languages: English
• ISSNs: 1868-1743

Abstract

Abstract: Here, we report two‐class classification models for organic molecules ("ligands") able to bind various metal cations in water. The modeling was performed on 30 data sets, each corresponding to a particular metal, using the Naïve Bayes method and the ISIDA fragment descriptors. The ligands were classified on weak and strong binders according to threshold of the logarithm of the stability constant of the 1 : 1 (metal : ligand) complexes. The "consensus models" consisted each of 50 best individual models demonstrated a good predictive performance in 5‐fold cross validation: the balanced accuracy ( BA ) varies from 0.965 (Yb 3+ ) to 0.767 (Mg 2+ ). The best predictions ( BA >0.90) were obtained for the binders of rare‐earth metals (Yb 3+, Tm 3+, Er 3+, Lu 3+, Ho 3+, Gd 3+ and Dy 3+ ) and Pb 2+ . For 17 small external test sets of new ligands, BA varies from 0.800 to 1. The impact of variables selection on the predictive performance of the models is discussed. Abstract :