Complementarity of the residue-level protein function and structure predictions in human proteins

Area of Science:

• Computational Biology and Bioinformatics
• Structural Bioinformatics
• Molecular Biophysics

Background:

• Sequence-based predictors analyze protein characteristics like intrinsic disorder, secondary structure, solvent accessibility, and nucleic acid binding.
• Existing studies often evaluate predictors for single characteristics in isolation, despite their application to the same proteins.
• A gap exists in understanding the complementarity between predictors targeting diverse residue-level protein characteristics.

Purpose of the Study:

• To investigate the complementarity of diverse residue-level protein predictors.
• To bridge the gap between structure-trained and disorder-trained binding residue predictors.
• To assess if diverse predictors accurately reflect experimental relationships between protein characteristics.

Main Methods:

• Utilized a large, taxonomically consistent, low-similarity dataset of human proteins.
• Empirically combined structure-trained and disorder-trained predictors for DNA-binding and RNA-binding residues.
• Investigated the concordance of predictions with experimental data regarding secondary structure, solvent accessibility, interaction sites, and intrinsic disorder.

Main Results:

• Combining structure-trained and disorder-trained predictors significantly improved the predictive quality of DNA- and RNA-binding residues.
• Diverse residue-level predictions accurately reproduced known relationships between secondary structure, solvent accessibility, interaction sites, and intrinsic disorder.
• Empirical analysis confirmed that predictions align with experimentally observed correlations among these protein features.

Conclusions:

• Integrating diverse residue-level predictors offers substantial benefits for predicting protein function and structure.
• The complementarity of structure-based and disorder-based approaches enhances the accuracy of binding site predictions.
• This study supports the combined use of various residue-level predictors for a more comprehensive understanding of protein characteristics.