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

Liquid-liquid phase separation (LLPS) has emerged as a fundamental mechanism underlying the formation and regulation of membraneless cellular compartments and is increasingly implicated in diverse physiological processes and diseases. Alongside rapid experimental and high-throughput advances, bioinformatics data resources and computational models have expanded substantially, enabling systematic cataloguing of LLPS-associated components and prediction of phase-separation behavior from molecular features. However, the resulting computational landscape remains highly fragmented. In this review, we provide a comprehensive and critical synthesis of bioinformatics resources and predictive modelling approaches for LLPS. We examine and compare major LLPS databases, highlighting differences in evidence types, curation strategies, coverage, and cross-resource inconsistencies that limit integrative analysis. We then survey computational models across core LLPS prediction tasks, encompassing more than 40 representative algorithms and tracing methodological evolution from classical machine learning to deep learning and large language model-based frameworks. By integrating these advances, we identify a fundamental mismatch between molecule-centric data abstractions and the inherently multicomponent, context-dependent organization of LLPS phenomena. We argue that future progress may benefit from event-centric frameworks that explicitly represent molecular assemblies, contextual conditions and observable phase behaviors, thereby providing a coherent foundation for next-generation LLPS datasets and computational models with improved mechanistic interpretability and translational relevance.