One bead per residue can describe all-atom protein structures

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• High-resolution biomolecular structures typically require atomistic detail.
• Experimental data and computational coarse-grained models often provide lower-resolution structural information.
• Bridging the resolution gap is crucial for understanding biological systems.

Purpose of the Study:

• To investigate the efficacy of machine learning for reconstructing atomistic models from reduced representations.
• To determine the minimal representation required for accurate all-atom structure generation.
• To develop rapid protocols for generating atomistic detail from low-resolution data.

Main Methods:

• Utilized advanced machine learning networks for model reconstruction.
• Employed a single bead per amino acid residue as the reduced representation.
• Developed a rapid, deterministic protocol for all-atom reconstruction from cryo-electron microscopy (cryo-EM) densities.

Main Results:

• Achieved accurate and stereochemically realistic all-atom protein structures from a single bead per residue representation.
• Demonstrated minimal information loss during reconstruction from reduced models.
• Successfully generated accurate models from cryo-EM densities, closely matching experimental structures.

Conclusions:

• Simplified protein representations (e.g., one bead per residue) are sufficient for generating high-quality atomistic models using machine learning.
• Machine learning frameworks can effectively encode structural knowledge, enabling accurate reconstruction from lower-resolution data.
• This approach facilitates rapid addition of atomistic detail to experimental and computational low-resolution structures, advancing multi-scale modeling.