Information theory applied to the sparse gene ontology annotation network to predict novel gene function

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Many gene products have functions that are not well-understood, hindering biological research.
• Current in silico methods for predicting gene functions often rely on experimental data and struggle with sparsely annotated Gene Ontology (GO) terms.

Purpose of the Study:

• To develop a novel, highly accurate computational method for predicting molecular functions of genes.
• To address the limitations of existing methods in annotating sparsely characterized GO terms.

Main Methods:

• Developed and applied an information theory-based semantic similarity (ITSS) algorithm.
• Utilized existing Gene Ontology (GO) annotations for prediction.
• Performed 10-fold cross-validation and historical rollback validation for accuracy assessment.

Main Results:

• ITSS achieved high prediction accuracy (97% precision, 77% recall) on densely annotated GO data, comparable to other machine learning algorithms.
• ITSS demonstrated superior performance on sparsely annotated GO data, generating an order of magnitude more predictions.
• Historical rollback validation estimated a minimum precision of 51% for human gene annotations from 2003.

Conclusions:

• The ITSS algorithm offers a significant advancement in predicting molecular functions for genes, particularly those with limited existing annotations.
• This method has the potential to accelerate functional genomics research by providing accurate, automated gene function predictions.