Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery

Autores
Schottlender, Gustavo Ezequiel; Prieto, Juan Manuel; Marti, Marcelo Adrian; Fernández Do Porto, Darío Augusto
Año de publicación
2025
Idioma
inglés
Tipo de recurso
artículo
Estado
versión publicada
Descripción
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, EF2%), 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.
Fil: Schottlender, Gustavo Ezequiel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; Argentina
Fil: Prieto, Juan Manuel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina
Fil: Marti, Marcelo Adrian. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales; Argentina
Fil: Fernández Do Porto, Darío Augusto. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; Argentina. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina
Materia
BIOACTIVITY SIMILARITY INDEX
BSI
MACHINE LEARNING
MOLECULAR EMBEDDING BASELINES
ChemBERTa
CLAMP
VIRTUAL SCREENING
Nivel de accesibilidad
acceso abierto
Condiciones de uso
https://creativecommons.org/licenses/by-nc-sa/2.5/ar/
Repositorio
CONICET Digital (CONICET)
Institución
Consejo Nacional de Investigaciones Científicas y Técnicas
OAI Identificador
oai:ri.conicet.gov.ar:11336/291514
Acceder
id CONICETDig_8e52a00c53754fd5d34d92a96a8497d3
oai_identifier_str oai:ri.conicet.gov.ar:11336/291514
network_acronym_str CONICETDig
repository_id_str 3498
network_name_str CONICET Digital (CONICET)
spelling Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discoverySchottlender, Gustavo EzequielPrieto, Juan ManuelMarti, Marcelo AdrianFernández Do Porto, Darío AugustoBIOACTIVITY SIMILARITY INDEXBSIMACHINE LEARNINGMOLECULAR EMBEDDING BASELINESChemBERTaCLAMPVIRTUAL SCREENINGhttps://purl.org/becyt/ford/1.4https://purl.org/becyt/ford/1Structural 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, EF2%), 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.Fil: Schottlender, Gustavo Ezequiel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; ArgentinaFil: Prieto, Juan Manuel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; ArgentinaFil: Marti, Marcelo Adrian. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales; ArgentinaFil: Fernández Do Porto, Darío Augusto. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; Argentina. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; ArgentinaFrontiers Media2025-11info:eu-repo/semantics/articleinfo:eu-repo/semantics/publishedVersionhttp://purl.org/coar/resource_type/c_6501info:ar-repo/semantics/articuloapplication/pdfapplication/pdfapplication/pdfapplication/pdfhttp://hdl.handle.net/11336/291514Schottlender, Gustavo Ezequiel; Prieto, Juan Manuel; Marti, Marcelo Adrian; Fernández Do Porto, Darío Augusto; Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery; Frontiers Media; Frontiers in Bioinformatics; 5; 11-2025; 1-172673-7647CONICET DigitalCONICETenginfo:eu-repo/semantics/altIdentifier/url/https://www.frontiersin.org/articles/10.3389/fbinf.2025.1695353/fullinfo:eu-repo/semantics/altIdentifier/doi/10.3389/fbinf.2025.1695353info:eu-repo/semantics/openAccesshttps://creativecommons.org/licenses/by-nc-sa/2.5/ar/reponame:CONICET Digital (CONICET)instname:Consejo Nacional de Investigaciones Científicas y Técnicas2026-08-25T15:03:25Zoai:ri.conicet.gov.ar:11336/291514instacron:CONICETInstitucionalhttp://ri.conicet.gov.ar/Organismo científico-tecnológicoNo correspondehttp://ri.conicet.gov.ar/oai/requestdasensio@conicet.gov.ar; lcarlino@conicet.gov.arArgentinaNo correspondeNo correspondeNo correspondeopendoar:34982026-08-25 15:03:26.149CONICET Digital (CONICET) - Consejo Nacional de Investigaciones Científicas y Técnicasfalse
dc.title.none.fl_str_mv Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
title Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
spellingShingle Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
Schottlender, Gustavo Ezequiel
BIOACTIVITY SIMILARITY INDEX
BSI
MACHINE LEARNING
MOLECULAR EMBEDDING BASELINES
ChemBERTa
CLAMP
VIRTUAL SCREENING
title_short Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
title_full Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
title_fullStr Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
title_full_unstemmed Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
title_sort Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery
dc.creator.none.fl_str_mv Schottlender, Gustavo Ezequiel
Prieto, Juan Manuel
Marti, Marcelo Adrian
Fernández Do Porto, Darío Augusto
author Schottlender, Gustavo Ezequiel
author_facet Schottlender, Gustavo Ezequiel
Prieto, Juan Manuel
Marti, Marcelo Adrian
Fernández Do Porto, Darío Augusto
author_role author
author2 Prieto, Juan Manuel
Marti, Marcelo Adrian
Fernández Do Porto, Darío Augusto
author2_role author
author
author
dc.subject.none.fl_str_mv BIOACTIVITY SIMILARITY INDEX
BSI
MACHINE LEARNING
MOLECULAR EMBEDDING BASELINES
ChemBERTa
CLAMP
VIRTUAL SCREENING
topic BIOACTIVITY SIMILARITY INDEX
BSI
MACHINE LEARNING
MOLECULAR EMBEDDING BASELINES
ChemBERTa
CLAMP
VIRTUAL SCREENING
purl_subject.fl_str_mv https://purl.org/becyt/ford/1.4
https://purl.org/becyt/ford/1
dc.description.none.fl_txt_mv 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, EF2%), 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.
Fil: Schottlender, Gustavo Ezequiel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; Argentina
Fil: Prieto, Juan Manuel. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina
Fil: Marti, Marcelo Adrian. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Química Biológica de la Facultad de Ciencias Exactas y Naturales; Argentina
Fil: Fernández Do Porto, Darío Augusto. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Instituto de Calculo. - Consejo Nacional de Investigaciones Científicas y Técnicas. Oficina de Coordinación Administrativa Ciudad Universitaria. Instituto de Calculo; Argentina. Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Química Biológica; Argentina
description 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, EF2%), 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.
publishDate 2025
dc.date.none.fl_str_mv 2025-11
dc.type.none.fl_str_mv info:eu-repo/semantics/article
info:eu-repo/semantics/publishedVersion
http://purl.org/coar/resource_type/c_6501
info:ar-repo/semantics/articulo
format article
status_str publishedVersion
dc.identifier.none.fl_str_mv http://hdl.handle.net/11336/291514
Schottlender, Gustavo Ezequiel; Prieto, Juan Manuel; Marti, Marcelo Adrian; Fernández Do Porto, Darío Augusto; Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery; Frontiers Media; Frontiers in Bioinformatics; 5; 11-2025; 1-17
2673-7647
CONICET Digital
CONICET
url http://hdl.handle.net/11336/291514
identifier_str_mv Schottlender, Gustavo Ezequiel; Prieto, Juan Manuel; Marti, Marcelo Adrian; Fernández Do Porto, Darío Augusto; Beyond Tanimoto: a learned bioactivity similarity index enhances ligand discovery; Frontiers Media; Frontiers in Bioinformatics; 5; 11-2025; 1-17
2673-7647
CONICET Digital
CONICET
dc.language.none.fl_str_mv eng
language eng
dc.relation.none.fl_str_mv info:eu-repo/semantics/altIdentifier/url/https://www.frontiersin.org/articles/10.3389/fbinf.2025.1695353/full
info:eu-repo/semantics/altIdentifier/doi/10.3389/fbinf.2025.1695353
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
https://creativecommons.org/licenses/by-nc-sa/2.5/ar/
eu_rights_str_mv openAccess
rights_invalid_str_mv https://creativecommons.org/licenses/by-nc-sa/2.5/ar/
dc.format.none.fl_str_mv application/pdf
application/pdf
application/pdf
application/pdf
dc.publisher.none.fl_str_mv Frontiers Media
publisher.none.fl_str_mv Frontiers Media
dc.source.none.fl_str_mv reponame:CONICET Digital (CONICET)
instname:Consejo Nacional de Investigaciones Científicas y Técnicas
reponame_str CONICET Digital (CONICET)
collection CONICET Digital (CONICET)
instname_str Consejo Nacional de Investigaciones Científicas y Técnicas
repository.name.fl_str_mv CONICET Digital (CONICET) - Consejo Nacional de Investigaciones Científicas y Técnicas
repository.mail.fl_str_mv dasensio@conicet.gov.ar; lcarlino@conicet.gov.ar
_version_ 1874775082031644672
score 13.265058

Publicaciones similares