קולוקוויום בית הספר למדעי המחשב
"Machine learning for Structural Biology: from function to design"
Speaker: Jerome Tubiana (Ecole Normale Supérieure)
Room 420 | Checkpoint Building
Structural biology, the study of three-dimensional structures of biomolecules and their link to function, is undergoing a revolution stirred by machine learning. Indeed, machine learning algorithms such as the celebrated AlphaFold enable structure prediction at scale without actual simulation of the physical folding process. However, harnessing structures for function or interaction prediction remains an open problem. One challenge is that biomolecular structures are peculiar data modalities distinct from classical data modalities such as grids, graphs, or point clouds. Here I will present a geometric deep learning model tailored for protein structures and discuss applications to functional site annotation. A second challenge is that function is not always well-defined in terms of biochemical properties. I will present an unsupervised generative modeling approach for design of inhibitory peptides.