MSBack: Multiscale Backmapping of Highly Coarse-Grained Proteins Using Constrained Diffusion

Area of Science:

• Biophysics
• Computational Biology
• Structural Biology

Background:

• Coarse-grained (CG) molecular dynamics simulates biomolecular behavior but loses structural detail.
• Backmapping CG models to all-atom resolution is challenging, especially at low resolutions (less than one site per residue).
• Low-resolution CG models are crucial for simulating large complexes like viruses (e.g., SARS-CoV-2, HIV-1).

Purpose of the Study:

• To develop a method for accurate all-atom reconstruction from highly coarse-grained protein structures.
• To address the unmet challenge of backmapping proteins at resolutions below one site per residue.
• To enable detailed structural analysis of large biomolecular complexes.

Main Methods:

• Proposed MSBack, a diffusion model approach for backmapping.
• Constrained the diffusion process to fit all-atom coordinates to CG coordinates.
• Utilized perturbation of existing all-atom structures without retraining.
• Integrated physics-based methods for fine-grained backmapping.

Main Results:

• MSBack stochastically generates alpha-carbon traces matching CG coordinates.
• Successfully demonstrated full backmapping of a mature HIV-1 capsid.
• Achieved 1 Å resolution for the backmapped HIV-1 capsid structure.
• The method is applicable to proteins simulated at resolutions necessary for large complexes.

Conclusions:

• MSBack effectively reconstructs all-atom protein structures from low-resolution CG models.
• This method overcomes limitations in current backmapping techniques.
• Enables detailed structural insights into large biomolecular systems and their interactions.