Design of phase-separating biosystem via joint diffusion and positive-unlabeled guidance

Peiran Jiang, Adrita Das, Weifeng Wu, Simran Sodhi, Huaiying Zhang, Jose Lugo-Martinez

Keywords: Diffusion model, protein, RNA, phase separation, biocondensate design, Positive-unlabeled learning

Abstract:

Liquid-liquid phase separation (LLPS) is a widespread mechanism by which cells
organize their internal environment, leading to the formation of
membraneless organelles and biomolecular condensates. The ability to design
to generate synthetic condensates properties has significant implications for
biotechnology and medicine. While current studies focused on protein-driven
phase separation, it is now clear that many condensates are multi-component
systems, including proteins, DNA, and RNA. Hereby, We propose a joint
diffusion framework that leverages the compositional generative process in
and guide a joint generation with a positive-unlabeled (PU)
assumption. Experiments on our newly curated multi-component
phase separation dataset demonstrate the efficacy of generating unseen biosystems
with desired phase behaviors. Our approach demonstrates the feasibility of
\textit{in silico} designing and engineering multi-component phase-separating biosystems.

Presenter: ~Peiran_Jiang1

Format: Maybe: the presenting author will attend in person, contingent on other factors that still need to be determined (e.g., visa, funding).

Funding: No, the presenting author of this submission does not fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 80