Mining alpha-helix-forming molecular recognition features with cross species sequence alignments.

Area of Science:

• Computational Biology
• Biophysics
• Bioinformatics

Background:

• Molecular recognition elements (MoREs), also known as molecular recognition features (MoRFs), are crucial for protein-protein interactions and involve disorder-to-order transitions.
• Existing algorithms for predicting alpha-helix-forming MoRFs (alpha-MoRFs) were limited by small training datasets.

Purpose of the Study:

• To improve the accuracy of alpha-MoRF prediction algorithms.
• To develop a more robust predictor by expanding training data and integrating diverse predictive attributes.

Main Methods:

• Augmented the positive training set with additional alpha-MoRF examples and cross-species homologues.
• Established a negative training set using monomer structure chains from the Protein Data Bank (PDB).
• Incorporated attributes from disorder predictors, secondary structure predictions, and amino acid indices, utilizing a conditional probability method to select top features for neural network development.

Main Results:

• Developed alpha-MoRF-PredII, a neural network-based predictor.
• Achieved high prediction performance with sensitivity, specificity, and accuracy of 0.87 +/- 0.10, 0.87 +/- 0.11, and 0.87 +/- 0.08, respectively, over 10 cross-validations.
• Identified key attributes, including VSL2 and VL3 disorder predictions and amino acid physicochemical propensities, that significantly contribute to prediction accuracy.

Conclusions:

• The enhanced training data and attribute selection significantly improved alpha-MoRF prediction accuracy.
• alpha-MoRF-PredII represents a substantial advancement in identifying regions involved in protein-protein interactions through disorder-to-order transitions.
• The study highlights the potential for further improvements in predicting intrinsically disordered protein regions and their roles in molecular recognition.