Unveiling hidden intermediate states in protein folding with AI-based conditional transition clustering

Area of Science:

• Computational biophysics
• Biomolecular simulations
• Artificial intelligence in science

Background:

• Protein folding mechanisms and intermediate states are crucial but challenging to study.
• Molecular dynamics (MD) simulations generate large datasets, but extracting kinetic models is difficult.
• Conventional methods like Markov state models have limitations in identifying conformational states.

Purpose of the Study:

• To present a novel AI-Based conditional transition clustering (CTC) framework for analyzing MD trajectories.
• To overcome limitations of state-centric methods by adopting a dynamics-centric approach.
• To enable objective identification of protein conformational states and folding pathways.

Main Methods:

• AI-Based conditional transition clustering (CTC) framework.
• Utilizing AI-based normalizing flows to estimate conditional transition probabilities.
• Defining conformational states as kinetically trapped regions identified from system dynamics.

Main Results:

• CTC successfully identifies critical intermediate and transition states in protein folding simulations.
• The framework reveals protein folding pathways without prior assumptions on the number or properties of states.
• Identified states as 'kinetic islands' with low escape probabilities.

Conclusions:

• CTC offers a more objective and physically grounded method for analyzing complex biomolecular systems.
• The dynamics-centric approach enhances the discovery of protein conformational mechanisms.
• This AI-driven framework advances the study of protein folding dynamics.