Seminar by Dr. Kieran Didi

Oxford University, UK

Generative Modelling meets Sampling: All-atom protein design and the role of simulations

Generative modelling has been soaring in popularity in the last years, fuelled by advances in disciplines like natural language processing, computer vision and biology. In this talk, I will talk about some of our efforts about efficient and performant protein design (including our Proteina project and recent extensions), as well as the limitations in this space when it comes to data. Following on this, I describe our efforts to remedy this via simulation (such as in the MISATO project) as well as recent developments in training normalising flows that might enable further scaling of these simulations beyond approaches like Multimap-TFEP.