Zoe Cournia
Distinguished by ISQBP with the Gilda Loew Lectureship
Affiliation
Biomedical Research Foundation, Academy of Athena, Greece - Research
Allostery in Drug Discovery: From MD to ML
Abstract
Allosteric regulation is a fundamental biological mechanism that can control critical cellular processes via allosteric modulator binding to protein distal functional sites. These protein allosteric sites are increasingly targeted in drug discovery because they can enable selective and specific modulation with fewer side effects [1]. However, predicting protein allosteric mechanisms and binding sites remains challenging due to the limited available data and the inherent complexity of allostery [2]. In this talk, we will discuss using Molecular Dynamics simulations and Machine Learning to uncover allosteric mechanisms of membrane proteins [3,4] and to target protein-membrane interfaces for allosteric drug design. As molecular simulations require abundant computational resources to identify protein-protein and protein-membrane interfaces, we describe an ensemble machine learning methodology to predict protein-membrane interfaces of peripheral membrane proteins [5] and present a drug design pipeline for drugging protein-membrane interfaces using the DREAMM (Drugging pRotein mEmbrAne Machine learning Method) web-server https://dreamm.ni4os.eu.[6] To support the development of allosteric site prediction models, we assemble an updated database of over 3,000 allosteric sites in protein structures, integrating multiple sources of annotations across diverse protein families [7]. Using these data, we develop AlloPockets, a deep learning model that predicts allosteric sites using comprehensive protein descriptors derived from sequence, structure, and dynamics (0.66 MCC, 0.83 F1 score). AlloPockets outperforms existing machine learning allosteric site prediction tools in a benchmark of 10 other allosteric prediction tools. Feature importance analysis indicates that AlloPockets captures hydrophobicity and complex dynamical properties that enable robust identification of allosteric pockets across conformational states. Together, these findings highlight the potential utility of AlloPockets for prospective structure-based allosteric drug design. Data, code, and models for allosteric site prediction are available at https://github.com/zoecournia/AlloPockets. References
[1] Chatzigoulas and Cournia, WIREs Comput. Mol. Sci. 11 (2021).
[2] Nerín-Fonz,and Cournia, Curr. Opin. Struct. Biol. 85 (2024) 102774.
[3] Kotzampasi and Cournia, Comms Chem (2025)
[4] Kotzampasi et al, CSBJ (2024)
[5] Chatzigoulas and Cournia, Briefings in Bioinformatics, 23, bbab518 (2022)
[6] Chatzigoulas and Cournia, Bioinformatics, 38, 5449-5451 (2022)
[7] He et al. Nucleic Acids Res. 52 (2024) D376–D383.