Gene function prediction in five model eukaryotes exclusively based on gene relative location through machine...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Most gene functions remain unknown, hindering biological research.
• Current gene function prediction methods often rely on sequence data, but gene location is also a strong indicator of function.
• Existing location-based methods have limitations due to their reliance on sequence identity.

Purpose of the Study:

• To investigate the predictive power of gene location for function prediction in eukaryotes.
• To develop and evaluate machine learning models trained exclusively on genomic location features.
• To compare the performance of location-based models against sequence-based methods.

Main Methods:

• Utilized machine learning models trained solely on features derived from gene locations within genomes.
• Applied models to predict gene functions across five model eukaryotic species: yeast, C. elegans, D. melanogaster, M. musculus, and H. sapiens.
• Compared model performance against the BLAST algorithm for predicting gene functions.

Main Results:

• Gene location features alone were sufficient to predict thousands of gene functions across model eukaryotes.
• The developed location-based models outperformed BLAST in predicting terms within the Biological Process and Cellular Component Gene Ontology categories.
• Demonstrated that gene location can be a more effective predictor of function than sequence in certain contexts.

Conclusions:

• Genomic gene location is a significant and underutilized feature for automated gene function prediction.
• Location-based machine learning approaches offer a powerful alternative or complement to sequence-based methods.
• This study highlights the potential of leveraging genomic context for advancing functional genomics.