Âé¶¹ÒùÔº

July 18, 2025

New AI-powered method accelerates protein simulations and reveals complex folding dynamics

Conceptual illustration of the pipeline for building and testing a transferable, bottom–up, machine-learned, CG protein force field. Credit: Nature Chemistry (2025). DOI: 10.1038/s41557-025-01874-0
× close
Conceptual illustration of the pipeline for building and testing a transferable, bottom–up, machine-learned, CG protein force field. Credit: Nature Chemistry (2025). DOI: 10.1038/s41557-025-01874-0

An international team led by Einstein Professor Cecilia Clementi in the Department of Âé¶¹ÒùÔºics at Freie Universität Berlin has introduced CGSchNet, a machine-learned coarse-grained (CG) model that can accurately and efficiently simulate proteins like never before. The study is in the July 18, 2025, issue of Nature Chemistry.

Operating significantly faster than traditional all-atom , CGSchNet enables larger proteins and to be explored—offering potential applications in drug discovery and protein engineering that could advance cancer treatment methods, for example.

Developing a general CG model capable of capturing protein folding and dynamics has been a persistent challenge for scientists over the last 50 years. "This work is the first to demonstrate that can overcome this barrier and lead to a simulation system that approximates all-atom protein simulations without explicitly modeling solvent or atomic detail," says Prof. Clementi.

In CGSchNet, Prof. Clementi's team trained a graph to learn the effective interactions between the particles of the coarse protein simulation to reproduce the dynamics of a diverse set of thousands of all-atom simulations.

Unlike structure prediction tools, CGSchNet models the dynamical process, including intermediate states relevant to misfolding processes like the formation of amyloids, which are pathological protein aggregates that appear in cases of Alzheimer's disease, for example.

The model also simulates transitions between folded states—key to —and generalizes to proteins outside its training set, demonstrating strong chemical transferability. Moreover, it accurately predicts metastable states of folded, unfolded, and disordered proteins, which constitute the majority of biologically active proteins.

Such predictions were extremely difficult in the past due to the flexibility of these proteins. The model is also able to estimate the relative folding free energies of protein mutants, which previous simulation methods could not achieve due to computational limitations.

More information: Nicholas E. Charron et al, Navigating protein landscapes with a machine-learned transferable coarse-grained model, Nature Chemistry (2025). .

Journal information: Nature Chemistry

Load comments (0)

This article has been reviewed according to Science X's and . have highlighted the following attributes while ensuring the content's credibility:

fact-checked
peer-reviewed publication
trusted source
proofread

Get Instant Summarized Text (GIST)

CGSchNet, a machine-learned coarse-grained model, enables efficient and accurate protein simulations, capturing folding dynamics and intermediate states without explicit atomic detail. It generalizes to proteins beyond its training set, predicts metastable states, and estimates relative folding free energies, overcoming limitations of traditional molecular dynamics approaches.

This summary was automatically generated using LLM.