How structural proteomics is being advanced by artificial intelligence

Structural Proteomics by Artificial Intelligence: Understanding the formation of protein complexes is crucial in drug design and the development of therapeutic proteins such as antibodies. However, proteins can bind to each other in millions of different combinations, and current docking solutions used to predict these interactions can be very slow. Faster and more accurate solutions are needed to streamline the process.

In a preprint published earlier this year, a new machine learning model, EquiDock, was introduced that can quickly predict how two proteins will interact. Unlike other approaches, the model does not rely on a large sampling of candidates and has been shown to reach predictions 80 to 500 times faster than popular coupling software.

To learn more about EquiDock and how artificial intelligence (AI) methods are advancing the field of structural proteomics, Technology Networks spoke with the paper’s co-senior author, Octavian-Eugen Ganea, a postdoctoral researcher in the Science Laboratory of the Computing and Artificial Intelligence from MIT.

Molly Campbell (MC): For our readers who may not be familiar, can you describe your current research focus in proteomics?

Octavian Ganea (OG): My research uses AI (specifically, deep learning) to model aspects of molecules that are important in various applications, such as drug discovery.

Proteins are involved in most biological processes in our body. Two or more proteins with different functions interact and form larger machines, that is, complexes. They also bind to smaller molecules, such as those found in medications. These processes change the biological functions of individual proteins, for example, an ideal drug would inhibit a cancer-causing protein by binding to specific parts of its surface. I am interested in using deep learning to model these interactions and help and accelerate the research of chemists and biologists by providing better and faster computational tools.

MC: How are AI-based methods advancing in the field of proteomics, and specifically structural proteomics?

OG: Biological processes are inherently very complicated and have their own mysteries, even for experts on the subject. For example, to understand how interacting proteins bind to each other, humans or computers have to try all possible binding combinations to find the most plausible one. Intuitively, having two three-dimensional objects with very irregular surfaces, one has to rotate them and try to fit them in every possible way until one can find two complementary regions on both surfaces that would match very well in terms of their geometric and chemical patterns. This is a time-consuming process for both manual and computational approaches. Furthermore, biologists are interested in discovering new interactions across a very large set of proteins, such as the ~20,000-sized human proteome. This is important, for example, to automatically discover unexpected side effects of new treatments. Such a problem now becomes similar to an extremely large 3D puzzle in which one has to simultaneously scan the pieces to match, as well as understand how each pairwise join occurs by trying all possible combinations and rotations.

MC: Can you explain how you created EquiDock?

OG: EquiDock takes the 3D structures of two proteins and directly identifies which areas are likely to interact, which would otherwise be a tricky problem even for a biologist. Discovering this information is enough to understand how to rotate and orient the two proteins in their bound positions. EquiDock learns to capture complex docking patterns from a large set of approximately 41,000 protein structures using a geometrically constrained model with thousands of parameters that dynamically and automatically adjust until they solve the task very well.

MC: What are the potential applications of EquiDock?

OG: As already mentioned, EquiDock can enable rapid computational scanning for drug side effects. This goes along with large-scale virtual detection of drugs and other types of molecules (eg, antibodies, nanobodies, peptides). This is necessary to significantly reduce an otherwise infeasible astronomical search space for all of our current (even aggregated worldwide) experimental capabilities. A fast protein-to-protein docking method like EquiDock combined with a fast protein structure prediction model (like AlphaFold2 developed by DeepMind) would help in drug design, protein engineering, antibody generation or understanding of the mechanism of action of a drug, among many other interesting applications. critically needed in our search for better treatments for diseases.

Leave a Comment