GeoGraph: Geometric and Graph-based Ensemble Descriptors for Intrinsically Disordered Proteins

2025-10-01

AI Summary

Addressing the longstanding challenge of balancing physical interpretability and computational efficiency in predicting conformational ensembles of intrinsically disordered proteins (IDPs), this work introduces GeoGraph—a novel method that directly predicts ensemble-level statistical properties of residue contact maps from protein sequences. GeoGraph integrates coarse-grained molecular dynamics (CG-MD) simulations, geometry-aware graph neural networks (GNNs), and protein language models (PLMs). Crucially, it encodes physically grounded information—derived from CG-MD trajectories—into sequence-level and residue-level graph descriptors, enabling a surrogate model that is both physically principled and computationally efficient. Experiments demonstrate that GeoGraph significantly outperforms existing methods in predicting key biophysical observables—including radius of gyration (Rg), asphericity (Asph), and small-angle X-ray scattering (SAXS) profiles—while exhibiting strong generalization across diverse IDPs and maintaining rigorous physical consistency.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

While deep learning has revolutionized the prediction of rigid protein structures, modelling the conformational ensembles of Intrinsically Disordered Proteins (IDPs) remains a key frontier. Current AI paradigms present a trade-off: Protein Language Models (PLMs) capture evolutionary statistics but lack explicit physical grounding, while generative models trained to model full ensembles are computationally expensive. In this work we critically assess these limits and propose a path forward. We introduce GeoGraph, a simulation-informed surrogate trained to predict ensemble-averaged statistics of residue-residue contact-map topology directly from sequence. By featurizing coarse-grained molecular dynamics simulations into residue- and sequence-level graph descriptors, we create a robust and information-rich learning target. Our evaluation demonstrates that this approach yields representations that are more predictive of key biophysical properties than existing methods.

Problem

• Predict conformational ensembles of disordered proteins efficiently
• Overcome trade-off between physical grounding and computational cost
• Develop graph-based descriptors for residue contact prediction

Innovation

• Uses graph descriptors from molecular dynamics simulations
• Predicts ensemble-averaged contact-map topology from sequence
• Creates simulation-informed surrogate model for disordered proteins