Work

Small-Molecule Data Splitting for Machine Learning in Drug Design

Year: 2026

Type: preprint

Abstract: Data splitting is a critical determinant of machine-learning performance in small-molecule drug design. Different strategies impose distinct assumptions about structural similarity, temporal order, data provenance and chemical-space coverage, thereby affecting both evaluation realism and the interpretation of model performance. This perspective critically examines major splitting approaches, discusses their advantages, applicability and limitations, and offers evidence-based considerations for designing partitions aligned with intended prediction scenarios.

Source: ChemRxiv

DOI: 10.26434/chemrxiv.15009691/v1

Authors: Cao-Minh Truong, Van-Thinh To, Gia-Bao Truong, Alexey Orlov, Pedro J. Ballester, Olivier Taboureau, Viet-Khoa Tran-Nguyen

Institutions: Inserm, Université Paris Cité, Centre National de la Recherche Scientifique, Université de Strasbourg, Imperial College London

Language: English

Topic: Computational Drug Discovery Methods

Subfield: Drug Discovery

Field: Pharmacology, Toxicology and Pharmaceutics

Domain: Life Sciences

Open Access status: green

Funders: Agence Nationale de la Recherche, Engineering and Physical Sciences Research Council

Awards: Artificial Intelligence-Based Scoring Functions for Polypharmacology Prediction, Target-specific machine-learning scoring functions for reliable structure-based virtual screening

OpenAlex ID: W7214927326

Cites: 8

Cited by:

Related to: 10