SAnDReS 2.0 - Development of machine-learning models to explore the scoring function space

Aug 5, 2026

Upload by Violet

Classical scoring functions can deliver limited accuracy for predicting ligand binding affinity to proteins. With access to protein–ligand structures and affinity measurements, machine-learning models can be tailored to specific protein systems for improved predictive performance. SAnDReS presents a workflow integrating AutoDock Vina 1.2 and 54 Scikit-Learn regression methods to estimate binding affinity from structural inputs. The method supports models built from crystal, docked, and AlphaFold-generated structures, and outperforms classical scoring and several alternative machine-learning approaches in three case studies.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages2

Reading Time5 min

AccessFree to download

Abstract

Methodology and inputs

Proof of concept and case studies

Software availability

Keywords and supplementary materials

Journal Computational Chemistry Jun 20 2024 [https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449](https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449) SAnDR 1096-987XeS 2.0: Development of machine-learning models to explore the scoring function space Walter Filgueira de Azevedo Jr 1, Rodrigo Quiroga 2, Marcos Ariel Villarreal 2, Nelson José Freitas da Silveira 3, Gabriela Bitencourt-Ferreira 4, Amauri Duarte da Silva 5, Martina Veit-Acosta 6, Patricia Rufino Oliveira 7, Marco Tutone 8, Nadezhda Biziukova 9, Vladimir Poroikov 9, Olga Tarasova 9, Stéphaine Baud 10 Affiliations • 1Department of Physics, Institute of Exact Sciences, Federal University of Alfenas, Alfenas, Brazil. • 2Instituto de Investigaciones en Fisicoquímica de Córdoba (INFIQC), CONICET-Departamento de Química Teórica y Computacional, Facultad de Ciencias Químicas, Universidad Nacional de Córdoba, Ciudad Universitaria, Córdoba, Argentina. • 3Laboratory of Molecular Modeling and Computer Simulation, Federal University of Alfenas, Alfenas, Brazil. • 4Pontifical Catholic University of Rio Grande do Sul, Porto Alegre, Brazil. • 5Programa de Pós-Graduação em Tecnologias da Informação e Gestão em Saúde, Universidade Federal de Ciências da Saúde de Porto Alegre, Porto Alegre, Brazil. • 6Western Michigan University, Kalamazoo, Michigan, USA. • 7School of Arts, Sciences and Humanities, University of São Paulo, São Paulo, Brazil. • 8Dipartimento di Scienze e Tecnologie Biologiche Chimiche e Farmaceutiche (STEBICEF), Università di Palermo, Palermo, Italy. • 9Institute of Biomedical Chemistry, Moscow, Russia. • 10Laboratoire SiRMa, UMR CNRS/URCA 7369, UFR Sciences Exactes et Naturelles, Université de Reims Champagne-Ardenne, CNRS, MEDYC, Reims, France. Abstract Classical scoring functions may exhibit low accuracy in determining ligand binding affinity for proteins. The availability of both protein-ligand structures and affinity data make it possible to develop machine-learning models focused on specific protein systems with superior predictive performance. Here, we report a new methodology named SAnDReS that combines AutoDock Vina 1.2 with 54 regression methods available in Scikit-Learn to calculate binding affinity based on protein-ligand structures. This approach allows exploration of the scoring function space. SAnDReS generates machine-learning models based on crystal, docked, and AlphaFold-generated structures. As a proof of concept, we examine the performance of SAnDReS-generated models in three case studies. For all three cases, our models outperformed classical scoring functions. Also, SAnDReSgenerated models showed predictive performance close to or better than other machinelearning models such as KDEEP, CSM-lig, and ΔVinaRF20. SAnDReS 2.0 is available to download at [https://github.com/azevedolab/sandres](https://github.com/azevedolab/sandres). Keywords: binding affinity; crystal structure; machine learning; protein–ligand interactions; scoring function space. © 2024 Wiley Periodicals LLC. Este Artículo se encuentra embargado hasta el día 21 de junio 2025 En el siguiente enlace [https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449](https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449 se encuentra)[ se encuentra](https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449 se encuentra)[ ](https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.27449 se encuentra)[disponible:](disponible:) “Supporting Information” Supplementary Material S1 . Features used to build machine-learning models. Supplementary Material S2 . Regression methods. Supplementary Material S3 . Scoring functions[a] . Supplementary material S4 . Docking results for CDK2 structures with inhibitors bound to the ATP-binding pocket. Supplementary Material S5 . Predictive performance for poses of the test set (case study 1) . Supplementary Material S6 . Predictive performance for docked pose structures of the test set (case s

FAQs

What problem does SAnDReS address in ligand binding affinity prediction?

SAnDReS targets the limited accuracy of classical scoring functions when determining ligand binding affinity for proteins.

How does SAnDReS 2.0 compute binding affinity?

It combines AutoDock Vina 1.2 with 54 regression methods from Scikit-Learn to calculate binding affinity from protein–ligand structures.

Which types of structures are used to build and evaluate the machine-learning models?

Models are generated from crystal structures, docked structures, and AlphaFold-generated structures, then tested in three case studies.

Aug 5, 2026

Upload by Violet