Rethinking bioinformatics in liquid-liquid phase separation: data resources, predictive models, and an event-centric...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Liquid-liquid phase separation (LLPS) is crucial for forming membraneless cellular compartments and regulating physiological processes.
• LLPS is increasingly linked to various diseases, driving research in experimental and computational approaches.
• Bioinformatics resources and computational models for LLPS have grown but remain fragmented.

Purpose of the Study:

• To provide a comprehensive synthesis of current bioinformatics resources and predictive modeling approaches for LLPS.
• To critically examine and compare major LLPS databases and computational models.
• To identify limitations and propose future directions for LLPS research.

Main Methods:

• Systematic review and critical synthesis of existing LLPS bioinformatics databases.
• Survey and comparison of over 40 computational models for LLPS prediction, including machine learning and deep learning frameworks.
• Analysis of data abstractions and methodological evolution in LLPS modeling.

Main Results:

• Major LLPS databases exhibit inconsistencies in evidence types, curation, and coverage, hindering integrative analysis.
• Computational models for LLPS prediction show evolution from classical machine learning to advanced deep learning and large language models.
• A fundamental mismatch exists between molecule-centric data and the multicomponent, context-dependent nature of LLPS.

Conclusions:

• Future LLPS research requires a shift towards event-centric frameworks to capture molecular assemblies, context, and phase behaviors.
• Event-centric frameworks can provide a coherent foundation for next-generation LLPS datasets and models.
• Improved mechanistic interpretability and translational relevance are expected from event-centric approaches in LLPS research.