Tuning intrinsic disorder predictors for virus proteins

Area of Science:

• Virology
• Structural Biology
• Bioinformatics

Background:

• Many virus-encoded proteins contain intrinsically disordered regions (IDRs) essential for viral replication and immune evasion.
• The rapid increase in viral genomic data outpaces experimental structure determination, highlighting the need for accurate computational prediction methods.
• Existing disorder predictors are often validated on diverse protein sets, potentially limiting their accuracy for specific viral protein contexts.

Purpose of the Study:

• To evaluate the performance of various computational disorder prediction methods specifically on viral proteins.
• To compare the accuracy of these methods on viral versus non-viral protein datasets.
• To develop an improved prediction strategy by combining multiple predictor outputs.

Main Methods:

• Assessed the accuracy of 21 different protein disorder prediction tools on a curated set of 126 viral proteins.
• Compared the performance of these tools on viral proteins against their performance on non-viral proteins.
• Implemented a random forest classifier to create an ensemble model integrating outputs from individual predictors.

Main Results:

• Significant variations in prediction accuracy were observed among the 21 methods when applied to viral proteins.
• Disorder prediction performance differed between viral and non-viral protein datasets.
• The random forest ensemble approach demonstrated a substantial improvement in prediction accuracy, with a mean 36% increase in Matthews correlation coefficient.
• The ensemble predictor was successfully applied to SARS-CoV-2 ORF6, predicting its disordered nature and role in immune response inhibition.

Conclusions:

• Computational disorder prediction methods exhibit differential performance on viral proteins compared to non-viral ones.
• An ensemble approach, particularly using random forest classifiers, enhances the robustness and accuracy of viral protein disorder prediction.
• Accurate prediction of disordered regions in viral proteins is vital for understanding virus-host interactions and developing antiviral strategies.