NMA-tune: Generating Highly Designable and Dynamics Aware Protein Backbones

Urszula Julia Komorowska, Francisco Vargas, Alessandro Rondina, Pietro Lio, Mateja Jamnik

Abstract: Protein's backbone flexibility is a crucial property that heavily influences its functionality. Recent work in the field of protein diffusion probabilistic modelling has leveraged Normal Mode Analysis (NMA) and, for the first time, introduced information about large scale protein motion into the generative process. However, obtaining molecules with both the desired dynamics and designable quality has proven challenging. In this work, we present NMA-tune, a new method that introduces the dynamics information to the protein design stage. NMA-tune uses a trainable component to condition the backbone generation on the lowest normal mode of oscillation. We implement NMA-tune as a plug-and-play extension to RFdiffusion, show that the proportion of samples with high quality structure and the desired dynamics is improved as compared to other methods without the trainable component, and we show the presence of the targeted modes in the Molecular Dynamics simulations.

Lay Summary: One of the most promising directions for Generative AI applications in medicine is designing novel proteins. Those biomolecules are essential for many processes happening in nature, including processes within human bodies. Proteins' flexibility is a crucial property that heavily influences their exact functionality. In many cases the collective motions of proteins' building blocks can be explained by the Normal Mode Analysis (NMA), which is the theory of harmonic oscillations within a protein. Recent work in the field leveraged NMA to introduce information about large scale protein motion into the design process, which was the first step towards dynamics-informed design. However, obtaining molecules with both the desired dynamics and high structure quality that obeys physical constraints has proven challenging. In this work, we present NMA-tune, a new method that introduces the dynamics information to the protein design stage. NMA-tune uses a trainable component (a deep neural network) in order to condition the backbone generation on the lowest normal mode of oscillation. We implement NMA-tune as a plug-and-play extension to well-know generative model that was shown to generate realistic protein structures. We show that the proportion of samples with high quality structure and the desired dynamics is improved as compared to other methods without the trainable component, and we demostrate the presence of the targeted modes in the high quality, physics-based dynamics computer simulations.

Application-Driven Machine Learning: This submission is on Application-Driven Machine Learning.

Primary Area: Applications->Chemistry, Physics, and Earth Sciences

Keywords: generative models, diffusion models, protein design, conditional protein generation

Submission Number: 7769