New machine learning models revolutionize protein simulation

An international team led by Cecilia Clementi, professor of Einstein from the Faculty of Physics at the University of Berlin, presents a breakthrough in protein simulation. This study was published on July 18, 2025 Natural ChemistryWe present CGSCHNET, a coarse-grained (CG) model of machine learning that can accurately and efficiently simulate proteins like never before. CGSCHNET operates significantly faster than traditional all-atomic molecular dynamics, allowing large-scale proteins and complex systems to be investigated.

Developing common CG models that can capture protein folding and dynamics has been a permanent challenge for scientists over the past 50 years. “This work is the first to demonstrate that deep learning overcomes this barrier and leads to simulation systems that are closer to all-atomic protein simulations without explicitly modeling solvent or atom details,” Cecilia clementi said.

In CGSCHNET, Clementi's team trained graph neural networks to learn effective interactions between particles in coarse protein simulations to replicate the dynamics of all thousands of all-atomic simulations. Unlike structural prediction tools, CGSCHNET models dynamic processes that involve intermediate states associated with misfolding processes, such as the formation of amyloid, a pathological protein aggregate that appears in the case of Alzheimer's disease. This model also simulates transitions between folded states (keys of protein function), generalizes to proteins outside the training set, demonstrating strong chemical transfer potential. Additionally, it accurately predicts the metastub state of folded, expanded, and impaired proteins that make up the majority of biologically active proteins. Such predictions have been extremely difficult in the past due to the flexibility of these proteins. This model can also estimate the relative folding free energies of protein mutants. This is something previous simulation methods could not be achieved due to computational limitations.

Professor Cecilia clementi is a theoretical and computational biophysicist. She previously conducted research at the University of Berlin as a Visiting Einstein Fellow at the Collaborative Research Center, Membrane Survey – Molecular Mechanisms and Cellular Function and Scale Cascades of Complex Systems. She is also the first scientist to be permanently recruited to work in Berlin with her support as an Einstein Visiting Fellow. Before moving to Berlin in 2020, Clementi was a professor of chemistry and physics at Rice University in Houston, Texas. Her role at Freie Universität allows her to enhance her research in Berlin in theoretical and computer-aided biophysics and build a bridge between experimental biophysics and applied mathematics.