FunPredCATH: An ensemble method for predicting protein function using CATH.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The rapid increase in protein sequencing generates vast amounts of unannotated data.
• Experimental protein annotation is time-consuming and costly.
• Computational prediction of Gene Ontology (GO) terms is crucial for efficient functional characterization.

Purpose of the Study:

• To develop and evaluate an ensemble computational method for predicting GO terms for uncharacterized proteins.
• To improve the accuracy and efficiency of protein functional annotation.
• To leverage protein sequence and structural family information for prediction.

Main Methods:

• An ensemble approach combining three base predictors for GO term prediction (Biological Process, Cellular Component, Molecular Function).
• Training models on UniProtGOA data and utilizing CATH domain resources for protein family identification.
• Employing statistics-based and set-based scoring methods, including Set Intersection and Set Union, within CATH Functional Families (FunFams).
• Incorporating the FunFams-Plus predictor for uncharacterized proteins, as used in the CAFA3 challenge.

Main Results:

• The ensemble method, FunPredCATH, demonstrated strong performance against the CAFA3 benchmark and DomFun.
• FunPredCATH compared favorably with top-performing methods in the CAFA3 challenge across different ontologies.
• The ensemble approach consistently outperformed its individual base predictors.

Conclusions:

• The developed ensemble method provides a robust and accurate approach for predicting GO terms.
• FunPredCATH offers a valuable tool for accelerating the functional annotation of uncharacterized proteins.
• The study highlights the effectiveness of combining diverse prediction strategies and leveraging structural information for improved protein function prediction.