Empirical Assessment of Sequence-Based Predictions of Intrinsically Disordered Regions Involved in Phase Separation

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Genomics and Proteomics

Background:

• Phase separation drives the formation of membrane-less organelles, involving structured domains and intrinsically disordered regions (IDRs) in proteins.
• Over 30 computational tools exist to predict protein phase separation from sequences, with a focus on IDRs due to their disease relevance.
• Recent studies show accurate phase separation prediction for structured proteins, prompting evaluation of predictors for IDRs.

Purpose of the Study:

• To evaluate the performance of amino acid-level predictors for identifying phase-separating intrinsically disordered regions (IDRs).
• To compare predictor accuracy in scenarios including both structured and disordered regions versus disordered regions only.
• To identify the most accurate computational tool for predicting phase-separating IDRs.

Main Methods:

• Evaluated eight representative amino acid-level phase separation predictors.
• Utilized a well-annotated, low-similarity test dataset.
• Assessed performance under two complementary evaluation scenarios: including all sequences and focusing solely on disordered regions.

Main Results:

• Many predictors showed accuracy in the easier scenario (structured and disordered sequences).
• Modern disorder predictors matched phase separation predictors in differentiating phase-separating IDRs from structured regions.
• In the challenging scenario (disordered regions only), most predictors showed modest accuracy, with some exhibiting bias towards classifying disordered residues as phase-separating.

Conclusions:

• Disorder predictors underperform when predicting phase separation solely within disordered regions.
• PSPHunter demonstrated the highest accuracy for identifying phase-separating IDRs across both evaluation scenarios.
• The study highlights the need for robust predictors that can accurately identify phase-separating IDRs, especially in challenging contexts.