Protein function prediction using functional inter-relationship.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-throughput sequencing generates vast amounts of protein data, necessitating efficient function identification.
• Traditional methods and existing machine learning approaches struggle with predicting large sets of protein functions due to redundancy and inter-relationships.

Purpose of the Study:

• To develop a machine learning approach for predicting large sets of protein functions by utilizing inter-relationships between functions.
• To reduce redundancy among protein functions and improve prediction accuracy.

Main Methods:

• Employed statistical measures (Pearson's correlation coefficient, Jaccard similarity coefficient) to identify and remove redundant functions.
• Trained a machine learning model on a reduced, non-redundant set of functions.
• Evaluated the model using Direct Mapping and Ensemble approaches for inverse transformation of function sets.

Main Results:

• The proposed model successfully predicts protein functions, including specific functions missed by other methods.
• Demonstrated improved predictability by accounting for inter-relationships between functions.
• Achieved promising results on DeepGO and CAFA3 datasets using various feature and function sets.

Conclusions:

• Machine learning models that incorporate functional interrelationships can significantly enhance protein function prediction accuracy.
• The developed approach effectively addresses the challenge of predicting large and complex sets of protein functions.
• The methodology offers a robust solution for annotating newly discovered proteins with high-throughput sequencing data.