A large dataset curation and benchmark for drug target interaction

TL;DR

The work addresses the lack of standardized, scalable datasets for drug-target interaction (DTI) prediction by constructing a large, multi-source benchmark that integrates PubChem, BindingDB, and ChEMBL. It introduces a unified three-table representation (pairs, ligands, targets) and three split strategies (lenient, cold-ligand, cold-target) to enable robust evaluation across unseen entities. The authors demonstrate the benchmark with a pretrained Protein Language Model–based DTI predictor and a Morgan fingerprint–based ligand encoder, highlighting gains from focal loss and deeper architectures on a large dataset. This benchmark and its open-source code/data provide a scalable, reproducible foundation for advancing DTI methods in drug discovery and repurposing.

Abstract

Bioactivity data plays a key role in drug discovery and repurposing. The resource-demanding nature of \textit{in vitro} and \textit{in vivo} experiments, as well as the recent advances in data-driven computational biochemistry research, highlight the importance of \textit{in silico} drug target interaction (DTI) prediction approaches. While numerous large public bioactivity data sources exist, research in the field could benefit from better standardization of existing data resources. At present, different research works that share similar goals are often difficult to compare properly because of different choices of data sources and train/validation/test split strategies. Additionally, many works are based on small data subsets, leading to results and insights of possible limited validity. In this paper we propose a way to standardize and represent efficiently a very large dataset curated from multiple public sources, split the data into train, validation and test sets based on different meaningful strategies, and provide a concrete evaluation protocol to accomplish a benchmark. We analyze the proposed data curation, prove its usefulness and validate the proposed benchmark through experimental studies based on an existing neural network model.

Figures (1)

• Figure 1: Train and validation AUPR using cross entropy (CE) loss vs. focal loss