Exploiting ontology graph for predicting sparsely annotated gene function.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Predicting gene and protein function using molecular interaction networks aids in refining biological databases like Gene Ontology (GO).
• Sparsely annotated functions (<10 genes) present a significant challenge for prediction algorithms due to limited data, leading to overfitting and poor performance.
• Existing function prediction algorithms struggle with the 'overfitting' issue of sparsely annotated functions and scalability to large annotation catalogs.

Purpose of the Study:

• To develop a novel algorithm for predicting gene function that effectively handles sparsely annotated functions.
• To improve the scalability and accuracy of function prediction methods for large biological datasets.
• To address the overfitting problem inherent in predicting functions with limited annotated genes.

Main Methods:

• Proposed a novel function prediction algorithm named clusDCA.
• Implemented an information transfer mechanism between similar functional labels to mitigate overfitting.
• Ensured the method's scalability for datasets with a large number of annotations.

Main Results:

• clusDCA significantly outperformed state-of-the-art algorithms in predicting sparsely annotated functions across yeast, mouse, and human datasets.
• The method maintained high performance on labels with sufficient annotations.
• Demonstrated accurate prediction of genes for previously unannotated functional labels by leveraging ontology graph structure and related gene annotations.

Conclusions:

• clusDCA effectively addresses the challenge of predicting sparsely annotated functions by transferring information between similar labels.
• The algorithm is scalable and improves prediction accuracy without compromising performance on well-annotated functions.
• clusDCA demonstrates a robust approach to gene function prediction, highlighting the utility of inter-label similarity in biological networks.