flDPnn3: Fast and Accurate Prediction of Intrinsic Disorder in Protein Sequences

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Protein intrinsic disorder plays a crucial role in various biological processes.
• Accurate prediction of intrinsically disordered regions is essential for understanding protein function.
• Existing disorder prediction tools have limitations in speed, accuracy, or coverage.

Purpose of the Study:

• To develop and evaluate flDPnn3, a novel computational tool for predicting protein intrinsic disorder.
• To assess the performance of flDPnn3 against other state-of-the-art disorder predictors.
• To provide a fast, accurate, and comprehensive solution for whole-proteome disorder prediction.

Main Methods:

• flDPnn3 utilizes sophisticated sequence-derived profiles, including a modern protein language model and additional disorder functions.
• Independent evaluation was conducted on the Disorder-NOX dataset within the CAID3 (3rd Critical Assessment of protein Intrinsic Disorder prediction) framework.
• Performance metrics included predictive accuracy, speed, and prediction coverage across diverse protein sequences.

Main Results:

• flDPnn3 demonstrated predictive quality comparable to the best disorder predictors in CAID3, even on low-sequence-similarity subsets.
• Runtime analysis revealed flDPnn3 to be 3-8 times faster than similarly accurate predictors.
• flDPnn3 achieved 100% protein prediction coverage, outperforming some accurate tools that failed on certain proteins.

Conclusions:

• flDPnn3 represents a significant advancement in predicting protein intrinsic disorder, offering a superior balance of speed, accuracy, and coverage.
• The tool's performance in CAID3 confirms its position as a leading method for disorder prediction.
• flDPnn3 is suitable for large-scale, whole-proteome analyses and is accessible via a web server.