In-silico prediction of disorder content using hybrid sequence representation

Area of Science:

• Protein bioinformatics
• Computational biology
• Molecular genetics

Background:

• Intrinsically disordered proteins (IDPs) are vital for cellular processes and implicated in human diseases.
• Accurate estimation of protein disorder content is essential for understanding protein families, functions, and disease associations.
• Current methods relying on per-residue disorder predictions may inaccurately estimate overall disorder content.

Purpose of the Study:

• To develop a novel, accurate sequence-based predictor for protein disorder content.
• To overcome limitations of residue-level disorder prediction methods.
• To provide a tool for high-throughput annotation of protein disorder.

Main Methods:

• Developed DisCon, a predictor utilizing 29 custom descriptors aggregating sequence, evolutionary, and structural information.
• Employed a ridge regression model for predicting disorder content.
• Evaluated DisCon against ten modern disorder predictors on a diverse test dataset.

Main Results:

• DisCon achieved a low mean squared error (0.05) and high Pearson correlation (0.68) in disorder content prediction.
• Demonstrated statistically significant improvement over existing methods, particularly for proteins with low sequence identity to training data.
• Showcased the utility of DisCon's predictions for refining binary annotations of disordered residues.

Conclusions:

• DisCon offers a high-quality, accurate alternative for high-throughput disorder content annotation.
• The predictor enhances the accuracy of disordered residue identification.
• A web server for DisCon is publicly available for research use.