Cross-organism learning method to discover new gene functionalities.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate gene and protein function knowledge is crucial for understanding biological processes and developing therapies.
• Existing gene and protein annotation databases are often incomplete, contain errors, and lack sufficient high-quality curated information.
• Computational methods for predicting gene annotations with associated reliability are essential for advancing biomedical research.

Purpose of the Study:

• To develop a novel cross-organism learning approach for predicting gene functionalities.
• To leverage annotations from well-studied organisms to improve predictions in related, less-studied organisms.
• To enhance the accuracy and reliability of gene annotation discovery.

Main Methods:

• Proposed a cross-organism learning strategy utilizing known annotations from one organism to predict functions for another.
• Introduced a new problem representation and random perturbation of annotations to enable supervised learning.
• Trained prediction models on extensive data from a well-annotated organism for application to a target organism.

Main Results:

• Demonstrated significant improvements in cross-organism gene annotation prediction compared to single-organism approaches.
• Validated the effectiveness of annotation perturbation for enhancing prediction model training.
• Showcased that the evolutionary distance between organisms did not negatively impact prediction accuracy.

Conclusions:

• The developed cross-organism learning method reliably predicts novel gene functionalities with associated likelihood values.
• Predicted annotations are valuable for complementing existing databases and improving biomedical knowledge discovery.
• The approach accelerates the gene annotation curation process by prioritizing novel predictions.