Computational algorithms to predict Gene Ontology annotations.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene function annotations are crucial for biological research and experimental design.
• Existing annotation datasets, like those from the Gene Ontology Consortium, are often incomplete or contain errors.
• Manual curation of gene annotations is time-consuming and expensive.

Purpose of the Study:

• To develop computational tools for predicting novel gene annotations.
• To improve the efficiency and accuracy of gene function annotation processes.
• To support scientists in the gene annotation curation workflow.

Main Methods:

• Utilized latent semantic analysis (LSA) with Latent Semantic Indexing (LSI) and Probabilistic LSA (pLSA).
• Proposed a novel method, Semantic IMproved Latent Semantic Analysis (SI-LSA), incorporating a gene clustering step.
• Implemented annotation weighting schemes to enhance LSA algorithm performance.

Main Results:

• Tested LSA variants and weighted approaches on Gene Ontology annotations for *Bos taurus*, *Danio rerio*, and *Drosophila melanogaster*.
• Demonstrated the capability of the methods in predicting novel gene annotations.
• Showed that weighting procedures significantly improve prediction accuracy, with results varying by dataset size and algorithm.

Conclusions:

• Semantic IMproved Latent Semantic Analysis (SI-LSA) outperformed other methods.
• SI-LSA, particularly with a suitable weighting policy, effectively predicts numerous novel gene annotations.
• The developed computational tool assists scientists in the gene functional annotation curation process.