Overfitting in machine learning models for predicting partial atomic charges of drug-like molecules

Denis Vadimovich Zverev; P. A. Nikiforova; Arslan Ramilevich Shaimardanov; Dmitry Alexandrovich Shulga; Vladimir Alexandrovich Palyulin

doi:10.71267/mencom.7930

Overfitting in machine learning models for predicting partial atomic charges of drug-like molecules

Mendeleev Communications 2026

View at publisher

Authors: Denis Vadimovich Zverev, P. A. Nikiforova, Arslan Ramilevich Shaimardanov, Dmitry Alexandrovich Shulga, Vladimir Alexandrovich Palyulin

DOI: 10.71267/mencom.7930

Cited by: 0

References: 27

Abstract

Overfitting in machine learning models for partial atomic charges was investigated for highly heterogeneous datasets common in medicinal chemistry. Random forest and multilayer perceptron models were trained and validated on a specially clustered dataset of drug-like molecules. Analysis of standard quality metrics for reproducing RESP charges showed that the trained models exhibit no evidence of overfitting.

Fields of Study & Subjects

Computational Drug Discovery Methods Advanced Chemical Physics Studies Machine Learning in Materials Science

Publication details

Journal / source: Mendeleev Communications
Volume: 36
Issue: 3
Pages: 257–259
Publication date: 1 January 2026
Language: RU

References 27