Beyond Tanimoto: A Learned Bioactivity Similarity Index Enhances Ligand Discovery

Gustavo Schottlender¹Juan Manuel Prieto²Marcelo Adrian Marti^3*Dario Fernández Do Porto^1*

• ¹Instituto de Cálculo, Universidad de Buenos Aires Facultad de Ciencias Exactas y Naturales, Buenos Aires, Argentina
• ²LUCAI BIO, Dover, United States
• ³IQUIBICEN, Universidad de Buenos Aires Facultad de Ciencias Exactas y Naturales, Buenos Aires, Argentina

Structural similarity metrics like the Tanimoto Coefficient (TC) miss many functionally related compounds—indeed, 60% of similarly bioactive ligand pairs in ChEMBL 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 across Pfam-defined protein groups on dissimilar pairs, BSI not only outperforms TC but also surpasses modern molecular embedding baselines (ChemBERTa and 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 to protein families 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, EF₂%), with group-specific models delivering the best enrichment and BSI-Large remaining competitive. In a realistic virtual-screening-like scenario against the target 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.