A framework for conditional diffusion modelling with applications in motif scaffolding for protein design
Tuesday June 18th, 4-5pm EST | Kieran Didi, MSc, Simon Mathis, PhD student and Francisco Vargas, PhD student (Cambridge)
Abstract: Many protein design applications, such as binder or enzyme design, require scaffolding a structural motif with high precision. Generative modelling paradigms based on denoising diffusion processes emerged as a leading candidate to address this motif scaffolding problem and have shown early experimental success in some cases. In the diffusion paradigm, motif scaffolding is treated as a conditional generation task, and several conditional generation protocols were proposed or imported from the Computer Vision literature. However, most of these protocols are motivated heuristically, e.g. via analogies to Langevin dynamics, and lack a unifying framework, obscuring connections between the different approaches. In this work, we unify conditional training and conditional sampling procedures under one common framework based on the mathematically well-understood Doob's h-transform. This new perspective allows us to draw connections between existing methods and propose a new variation on existing conditional training protocols. We illustrate the effectiveness of this new protocol in both, image outpainting and motif scaffolding and find that it outperforms standard methods.
Preprint #1: https://arxiv.org/abs/2312.09236
Preprint #2: https://arxiv.org/abs/2406.01781
Kieran was a master's student at the University of Cambridge and recently joined industry. He's interested in solving problems by designing and engineering proteins.
Francisco is a final year PhD student at the University of Cambridge, focusing on topics at the interface of diffusions, stochastic control and transport.
Simon is a PhD student at Computer Science department at the University of Cambridge. He’s interested in designing and engineering enzymes.