Bioactivity foundation model for hit identification and hit-to-lead optimization from sequence data
- 1Department of Data Science and Artificial Intelligence, Monash University, VIC, Australia; Drug Discovery Biology, Monash Institute of Pharmaceutical Sciences,
Abstract
Learning from large-scale bioactivity archives is complicated by heterogeneous assay readouts and differences in data reliability. Here we show that a model that predicts probability distributions over binding affinities from sequence data opens model training to vast and diverse troves of heterogeneous bioactivity data. We develop PSICHIC+, a bioactivity foundation model trained on approximately 82 million protein-ligand records using a distributional objective that preserves quantitative resolution while incorporating variation in data reliability. Given a protein sequence and ligand SMILES, PSICHIC+ predicts binding affinity in milliseconds. In zero-shot virtual-screening benchmarks without target-specific training data, it improves enrichment over regression- and classification-based approaches and performs competitively with leading structure-based methods. In lab-in-the-loop campaigns, distribution-guided acquisition improves recovery of high-activity compounds under fixed screening budgets. Combining PSICHIC+ with Boltz-2 improves hit-to-lead affinity prediction over Boltz-2 alone, with mean correlation approaching that of FEP+ on the evaluated benchmarks, without requiring high-resolution protein structures. These results support distributional learning as a scalable approach to using heterogeneous bioactivity data for hit identification and hit-to-lead optimization.
Related
- Analyzing Genomic Foundation Models for Viral Sequence Identification
- A foundation model learns the sequence and functional grammar of fully human heavy-chain-only antibodies
- Phenolic Compounds from Ruta chalepensis Var. angustifolia as Potential HMG-CoA Reductase Inhibitors: Bioactivity-Guided Fractionation and Preliminary Identification of Active Metabolites
- A closed-loop language-model agent for target-specific multi-objective hit-to-lead optimization
- OmniTCR: a foundation model unifying T cell receptor recognition prediction and conditional sequence generation
BibTeX
@misc{koh2026bioactivity,
title = {{Bioactivity foundation model for hit identification and hit-to-lead optimization from sequence data}},
author = {H. Y. Koh and H. J. W. Koh and Y. Zheng and V. Naranbhai and D. F. Schmidt and J. Barthelemy and Y.-J. Chen and A. T. N. Nguyen and S. Pan and L. T. May and G. I. Webb},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.26.762100},
url = {https://www.biorxiv.org/content/10.64898/2026.10.26.762100v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Bioactivity foundation model for hit identification and hit-to-lead optimization from sequence data
AU - H. Y. Koh
AU - H. J. W. Koh
AU - Y. Zheng
AU - V. Naranbhai
AU - D. F. Schmidt
AU - J. Barthelemy
AU - Y.-J. Chen
AU - A. T. N. Nguyen
AU - S. Pan
AU - L. T. May
AU - G. I. Webb
PY - 2026
DA - 2026/10/26
PB - bioRxiv
DO - 10.64898/2026.10.26.762100
UR - https://www.biorxiv.org/content/10.64898/2026.10.26.762100v1
AB - Learning from large-scale bioactivity archives is complicated by heterogeneous assay readouts and differences in data reliability. Here we show that a model that predicts probability distributions over binding affinities from sequence data opens model training to vast and diverse troves of heterogeneous bioactivity data. We develop PSICHIC+, a bioactivity foundation model trained on approximately 82 million protein-ligand records using a distributional objective that preserves quantitative resolution while incorporating variation in data reliability. Given a protein sequence and ligand SMILES, PSICHIC+ predicts binding affinity in milliseconds. In zero-shot virtual-screening benchmarks without target-specific training data, it improves enrichment over regression- and classification-based approaches and performs competitively with leading structure-based methods. In lab-in-the-loop campaigns, distribution-guided acquisition improves recovery of high-activity compounds under fixed screening budgets. Combining PSICHIC+ with Boltz-2 improves hit-to-lead affinity prediction over Boltz-2 alone, with mean correlation approaching that of FEP+ on the evaluated benchmarks, without requiring high-resolution protein structures. These results support distributional learning as a scalable approach to using heterogeneous bioactivity data for hit identification and hit-to-lead optimization.
ER -