TIDGN: A Transfer Learning Framework for Predicting Interactions of Intrinsically Disordered Proteins with High...

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning

Background:

• Intrinsically disordered proteins (IDPs) interactions are vital for cellular processes like liquid-liquid phase separation (LLPS).
• Experimental and simulation methods for studying IDP interactions face challenges, including limited training data for machine learning approaches.
• Developing accurate predictive models for IDP interactions is crucial for advancing biological understanding.

Purpose of the Study:

• To develop a novel machine learning model for predicting intrinsically disordered protein (IDP) interactions.
• To address the challenge of data scarcity in training predictive models for IDP interactions.
• To enhance the understanding of both homotypic and heterotypic IDP interactions.

Main Methods:

• Proposed a transfer learning-based invariant geometric dynamic graph model (TIDGN).
• Constructed datasets for IDP monomer structures and interaction events using all-atom molecular dynamics (MD) simulations.
• Employed a pretraining task module for dynamic structural encoding and a downstream task module for interaction site prediction.

Main Results:

• The TIDGN model effectively predicts IDP interactions, demonstrating strong performance, particularly for heterotypic interactions.
• Transfer learning significantly improved model performance, mitigating issues related to limited training data.
• Feature ablation analysis confirmed the importance of invariant geometric graph features in the model's predictive power.

Conclusions:

• The integration of transfer learning and invariant geometric graph networks offers a promising solution for data scarcity in IDP interaction prediction.
• TIDGN provides a robust computational tool for studying IDP interactions, advancing research in areas like LLPS.
• This approach paves the way for more accurate and efficient prediction of protein interactions in complex biological systems.