LMoI - Biomolecular language models for protein design

What is the relationship between the structure, dynamics and function of proteins? Can we computationally predict and tailor new functions into proteins? These are some of the fundamental questions that I plan to address in my academic research program at the Faculty of Biomedical Sciences, Università della Svizzera Italiana (USI), Lugano, Switzerland. Proteins are remarkable biomolecular nano-machines that perform most essential processes needed for the survival of living organisms. The objective of protein design is to harness this powerful framework in order to engineer new functions into proteins for biomedical and biotechnological applications. Although computational approaches have permitted designing a wide range of new protein structures, they still fail at changing the function of proteins. Currently, no simple physical model permits the modeling of all the necessary disorder, flexibility and dynamics, required for function. The dramatic growth of structural databases for biomolecules would allow a data-driven approach to this problem. Therefore, I propose a Deep Learning project that will shed light and ultimately help narrow the gaps in our theoretical understanding of the functional relationships between sequence, structure and dynamics in proteins. In an innovative approach, I intend to tackle the design process of functional proteins as a language-modeling problem. First, a biomolecular language model will be developed for predicting and designing the interaction networks observed in protein fragments. Then, this approach will be extended for modeling the interactions between proteins and ligands. Finally, these biomolecular language models will be tested for the design of new biosensors from the family of ubiquitous sensory modules (PAS domains). My approach is orthogonal to methods currently used in protein design and will create a new methodology for the design of proteins with distinct catalytic, pharmaceutical and sensing properties. This will have a significant impact on the design of, e.g., antibody therapeutics, new enzymes and molecular probes.