Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
Frontiers in Bioinformatics · 28.11.2025 · 5
DOI: 10.3389/fbinf.2025.1695353 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Bioinformatics, online erschienen am 28.11.2025. Verfasst von Gustavo Schottlender und 3 weiteren Autor:innen. Beteiligte Einrichtung: Instituto de Cálculo, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, CONICET, Universidad de Buenos Aires, Buenos Aires, Arge…. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 77 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Structural similarity metrics such as the Tanimoto coefficient (TC) miss many functionally related compounds—indeed, 60% of similarly bioactive ligand pairs in the ChEMBL database show TC < 0.30, revealing a major blind spot that constrains ligand-based discovery. Our motivation is to overcome this blind spot and enable the recovery of structurally different yet functionally equivalent chemotypes that structure-based similarity fails to detect. Here, we introduce the bioactivity similarity index (BSI), a machine learning model that estimates the probability that two molecules bind the same or related protein receptors. Trained under leave-one-protein-out (LOPO) across Pfam-defined protein groups on dissimilar pairs, BSI not only outperforms TC but also surpasses modern molecular embedding baselines (ChemBERTa and contrastive language-molecule pre-training (CLAMP), using cosine similarity) across protein families. We further develop a cross-family model (BSI-Large) that, while slightly below group-specific models, generalizes better and can be fine-tuned with less data, consistently improving over models trained from scratch. In retrospective validation on new ChEMBL v35 data, BSI achieves strong early-retrieval performance (top 2% enrichment factor, EF 2% ), with group-specific models delivering the best enrichment, and BSI-Large remaining competitive. In a realistic virtual screening-like scenario against the target gene ADRA2B, the mean rank of the next active, given a known active, improves from 45.2 (TC) to 3.9 (BSI), with 54.9 for ChemBERTa and 28.6 for CLAMP. Altogether, BSI complements, rather than replaces, structure-based similarity and embedding-based comparisons, extending hit finding to remote chemotypes that are structurally dissimilar yet functionally equivalent. The code is available at https://github.com/gschottlender/bioactivity-similarity-index .
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Instituto de Cálculo, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, CONICET, Universidad de Buenos Aires, Buenos Aires, Argentina.
- LUCAI BIO, Dover, DE, United States.
- Departamento de Química Biológica, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires (FCEyN-UBA), Buenos Aires, Argentina.
- Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales (IQUIBICEN) CONICET, Pabellón 2 de Ciudad Universitaria, Buenos Aires, Argentina.
Zitieren
APA
Schottlender, G., Prieto, J. M., Marti, M. A., & Fernández Do Porto, D. (2025). Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery. Frontiers in Bioinformatics, 5. https://doi.org/10.3389/fbinf.2025.1695353
Vancouver
Schottlender G, Prieto JM, Marti MA, Fernández Do Porto D. Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery. Frontiers in Bioinformatics. 2025;5. doi:10.3389/fbinf.2025.1695353
BibTeX
@article{Schottlender2025,
author = {Schottlender, Gustavo and Prieto, Juan Manuel and Marti, Marcelo A and Fernández Do Porto, Dario},
title = {Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery},
journal = {Frontiers in Bioinformatics},
year = {2025},
volume = {5},
doi = {10.3389/fbinf.2025.1695353}
}
EUVIMED