Enhanced Protein Intrinsic Disorder Prediction Through Dual-View Multiscale Features and Multi-objective Evolutionary Algorithm

A research team led by Shaokuan Wang has introduced D2MOE, a novel AI architecture designed to solve a critical problem in computational biology: predicting intrinsically disordered regions (IDRs) in proteins. These flexible regions, which lack fixed 3D structures, are essential for cell signaling and represent promising targets for drug discovery, but their dynamic nature makes accurate prediction notoriously difficult. D2MOE addresses this by implementing a two-stage approach that first extracts "dual-view" features—combining evolutionary sequence patterns with deep semantic representations—across multiple scales to capture both local amino acid preferences and long-range structural information.

The second stage employs a sophisticated multi-objective evolutionary algorithm (MOEA) that automatically discovers the optimal architecture for fusing these diverse features. Unlike previous methods that relied on rigid, manually-designed fusion strategies, this algorithm co-evolves both discrete feature selections and continuous fusion weights, performing a global search to maximize predictive accuracy while maintaining model efficiency. This hybrid approach, merging the representational power of deep learning with the optimization strength of evolutionary computation, allowed D2MOE to consistently outperform existing state-of-the-art methods across three standard benchmark datasets. The result is a more robust computational tool that reduces reliance on expert manual design and provides researchers with higher-confidence predictions of protein disorder.