A new geometric machine learning method promises to accelerate precision drug development

(Press-News.org) Proteins are the foundation of all life we currently know. With their virtually limitless diversity, they can perform a broad variety of biological functions, from delivering oxygen to cells and acting as chemical messengers to defending the body against pathogens. Furthermore, most biochemical reactions are only possible thanks to enzymes, a special type of protein catalysts.

The molecular surface of proteins is the key to their function, such as docking small molecules or other proteins or driving chemical reactions. Much like a key fits only one lock and activates it, proteins often interact exclusively with a single molecular structure that precisely matches their surface.

This principle is exploited in drug development: drug molecules are designed to bind to specific proteins, altering their surface and, consequently, their behavior. The newly created "neo-surface" can in turn form novel interactions with other proteins. Molecules designed to bring together different proteins that otherwise would not interact are called “molecular glues,” and are a promising modality to treat diseases by inactivating or degrading proteins that cause disease.

New Proteins with a molecular fingerprint

A long-term collaboration of Michael Bronstein, scientific director of AITHYRA, the new Institute of the Austrian Academy of Sciences (ÖAW), with the team of Bruno Correia at the EPFL Laboratory for Immunoengineering and Protein Design has pioneered the use of geometric deep learning architecture called "Molecular Surface Interaction Fingerprinting" (MaSIF)1 to design new proteins2 with desired molecular surface properties.

In a new study3 published in Nature this week, the team applied MaSIF to proteins with bound drug molecules and showed that it can be used to design proteins that bind to these neo-surfaces.

“One of the key challenges of machine learning approaches is their generalization ability, or how well the method works on data never seen before,” explains Michael Bronstein. “One of the surprising and satisfying outcomes of our study is that a neural network trained on natural interactions between proteins generalizes very well to protein-ligand neo-surfaces never seen before. It seems that geometric descriptors of molecular surfaces extracted by our method are a sort of “universal language” for protein interactions.”

“The new approach allows us to design switchable protein interactions,” Bruno Correia says. “We can create new protein binders that interact with target proteins only in the presence of a small molecule. This opens a new avenue to precise dosing and control of biological drugs such as those used in oncological immunotherapies.”

Experiments validate virtual results

The researchers experimentally validated their novel protein binders against three drug-bound protein complexes containing the hormone progesterone, the FDA-approved leukemia drug Venetoclax, and the naturally occurring antibiotic Actinonin, respectively. The protein binders designed using MaSIF successfully recognized each drug-protein complex with high affinity. The team explains that this was possible because MaSIF is based on general surface features that apply to proteins and small molecules alike, so they were able to map the small molecule features onto the same descriptor space that MaSIF was trained on for proteins.

“MaSIF has a relatively small number of parameters – around 70,000 versus billions for large deep learning systems like ChatGPT,” explains PhD student and co-author Arne Schneuing, “This is possible because we use only key surface features, resulting in a high level of abstraction. In other words, we don’t give the system the full picture; only the part we think matters for solving the problem.”

Co-first author Anthony Marchand is excited about the prospects of the new approach. “Our idea was to engineer an interaction in which a small molecule causes two proteins to come together. Some approaches have focused on screening for such small molecules, but we wanted to design a novel protein that would bind to a defined protein-drug complex.” He believes that “such designed chemically induced protein interactions will have the potential to expand the sensing repertoire and the assembly of new synthetic pathways in engineered cells for innovative drug-controlled cell-based therapies.”

###

Cited Literature:

1: P. Gainza et al., Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning, Nature Methods (2020)
2: P. Gainza et al., De novo design of protein interactions with learned surface fingerprints, Nature (2023)
3: A. Marchand et al., Targeting protein-ligand neosurfaces with a generalizable deep learning tool, Nature (2024)

Pictures attached:

1 Michael Bronstein landscape © Natascha Unkart
2 Michael-Bronstein portrait © Natascha Unkart
3 Michael Bronstein (left) with Bruno Correia (right) © Charlie Harris

The Study “Targeting protein-ligand neosurfaces using a generalizable deep learning approach” was published in Nature on January 15, 2025. DOI: 10.1038/s41586-024-08435-4 https://www.nature.com/articles/s41586-024-08435-4

Authors: Anthony Marchand, Stephen Buckley, Arne Schneuing, Martin Pacesa, Pablo Gainza, Evgenia Elizarova, Rebecca M. Neeser, Pao-Wan Lee, Luc Reymond, Maddalena Elia, Leo Scheller, Sandrine Georgeon, Joseph Schmidt, Philippe Schwaller, Sebastian J. Maerkl, Michael Bronstein & Bruno E. Correia

Funding: This work was supported the Swiss National Science Foundation, the National Center of Competence in Research in Molecular Systems Engineering, the National Center of Competence in Research in Catalysis , a EPSRC Turing AI World-Leading Research Fellowship, Microsoft Research AI4Science, VantAI, Huawei Technologies Düsseldorf, Reprodivac, the H2020 Marie Sklodowska-Curie EPFL-Fellows and the “Peter und Traudl Engelhorn Stiftung”.

Michael Bronstein is the scientific director of AITHYRA. He will bring world-leading expertise in developing novel machine learning technique for biological applications as well as a track record of research commercialization and technological spinoffs and industrial experience. Michael Bronstein is also the DeepMind Professor of AI at the University of Oxford. His research interests lie in geometric deep learning, graph neural networks, 3D shape analysis, protein design, non-human species communication. Michael Bronstein was previously Head of Graph Learning Research at Twitter, a Professor at Imperial College London and held visiting appointments at Stanford, MIT, and Harvard. Michael Bronstein received his PhD from the Technion Israel Institute of Technology in 2007.

AITHYRA - Research Institute for Biomedical Artificial Intelligence aims to develop AI-first approaches to transform biological sciences. This approach will drive a biological revolution in the next decade, ultimately improving human health. AITHYRA will combine the best of academia, industry, and startups, bringing together experts in AI and life sciences. The new institute of the Austrian Academy of Sciences is based in Vienna, Europe’s Life Science hub, and financially supported by the Boehringer Ingelheim Foundation.

www.oeaw.ac.at/aithyra/
 

For further information please contact:

Stefan Bernhardt

PR & Communications Manager
Phone +43-1/40160-70 056
sbernhardt@cemm.at

Sven Hartwig

Leiter Öffentlichkeit & Kommunikation
Österreichische Akademie der Wissenschaften
Phone +43 1 51581-1331
sven.hartwig@oeaw.ac.at

AITHYRA

Research Institute for Biomedical Artificial Intelligence
of the Austrian Academy of Sciences
Dr. Ignaz Seipel Platz 2
1010 Vienna
www.oeaw.ac.at/aithyra/

END