PhaseOM: an integrated multi-task framework for phase separation analysis and key residue detection.

Authors

Liping Xu, Shengming Zhou, Zixu Ran, Xianmian Qin, Tian Liu, Quan Zou, Fuyi Li, Cangzhi Jia

Published in

Journal of advanced research. Aug 21, 2026. Epub Aug 21, 2026.

Abstract

Liquid-liquid phase separation (LLPS) drives cellular organization through scaffold proteins that initiate condensate formation and client proteins that partition into them. Current computational tools lack precision in differentiating these functional classes and identifying key intrinsically disordered regions (IDRs), limiting mechanistic understanding of LLPS in health and disease.
This study develops PhaseOM, a unified computational framework to: (1) differentiate scaffold versus client proteins; (2) identify LLPS-associated IDRs; and (3) pinpoint critical functional residues, enabling systematic analysis of LLPS determinants.
We constructed a multi-task predictive framework. Standard datasets were curated from drLLPS and PDB. PhaseOM integrates four specialized models: (1) a Graph Attention Network fusing ProtT5 sequence embeddings and ESMFold-predicted structural features for scaffold identification; (2) an ensemble classifier leveraging 22 optimized physicochemical and structural attributes for client discrimination; (3) a model based on 17 disorder-related features for IDR detection; and (4) a deep multi-layer perceptron with multi-head self-attention for key residue prediction. The performance of these models was thoroughly assessed using independent test datasets to ensure robustness and reliability.
PhaseOM outperformed Seq2Phase on independent tests, achieving scaffold AUC of 0.9954 and client AUC of 0.9474 (MCC 0.7514). It improved client AUC by 0.21 and scaffold AUC by >0.07 over baseline. Key findings reveal that scaffolds are enriched in tyrosine/arginine, facilitating multivalency, while clients exhibit expanded conformations with cysteine/histidine enrichment. Empirical validation on α-synuclein isoforms confirmed high-fidelity predictions, with residue-level IDR accuracy ranging from 76% to 91%. PhaseOM provides the first unified platform for systematic proteome-wide phase separation analysis.
PhaseOM provides the first unified platform for comprehensive LLPS analysis, revealing distinct mechanistic principles and enabling rational design of biomolecules with tunable phase behavior for therapeutic applications.

PMID:
42628918
Bibliographic data and abstract were imported from PubMed on 22 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.