Machine-learning-based methods to generate conformational ensembles of disordered proteins.

Area of Science:

• Computational biology
• Biophysics
• Machine learning

Background:

• Intrinsically disordered proteins (IDPs) lack stable 3D structures, existing as dynamic conformational ensembles.
• Traditional methods like molecular dynamics simulations to study IDP ensembles are computationally expensive.
• Machine learning (ML) offers a potential alternative for generating IDP conformational ensembles.

Purpose of the Study:

• To develop and validate a novel ML-based computational pipeline for generating conformational ensembles of intrinsically disordered proteins.
• To demonstrate a proof-of-principle for using ML to predict higher-dimensional properties of IDPs.

Main Methods:

• A two-stage ML pipeline was devised.
• Stage 1: Supervised ML models predicted 2D ensemble-derived properties from related sequences.
• Stage 2: Denoising diffusion models generated 3D coarse-grained conformational ensembles based on 2D predictions.

Main Results:

• The ML pipeline successfully predicted 2D properties and generated 3D conformational ensembles.
• Model accuracy was validated using multiple metrics.
• The approach was trained on a dataset of coarse-grained molecular dynamics simulations for synthetic sequences.

Conclusions:

• Machine learning techniques are applicable to predicting higher-dimensional properties of intrinsically disordered proteins.
• This ML pipeline provides a computationally efficient method for studying IDP conformational ensembles.
• This work opens new avenues for exploring the structural dynamics of disordered proteins using AI.