A postprocessing method in the HMC framework for predicting gene function based on biological instrumental data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene function prediction is a complex hierarchical multi-label classification (HMC) problem.
• Existing local approach methods often require preliminary results processing.
• Biological data analysis necessitates accurate hierarchical predictions.

Purpose of the Study:

• To propose a novel preliminary results processing method for HMC problems.
• To enhance gene function prediction accuracy by ensuring hierarchical consistency.
• To address the challenges of predicting gene function using biological instrumental data.

Main Methods:

• Introduced the "nodes interaction method" for revising preliminary HMC results.
• Phase 1: Utilized a Bayesian network to exploit label dependency and hierarchical interactions.
• Phase 2: Adjusted predictions to strictly adhere to hierarchy constraints.

Main Results:

• The nodes interaction method successfully revised preliminary predictions, ensuring hierarchical consistency.
• Implementing this method within an HMC framework improved gene function prediction performance.
• Demonstrated superior performance compared to state-of-the-art methods on yeast datasets annotated by Gene Ontology (GO).

Conclusions:

• The nodes interaction method is effective for improving HMC performance in gene function prediction.
• The method's ability to handle complex hierarchies, like the GO, is a significant advancement.
• This approach offers a promising solution for accurate and biologically consistent gene function prediction.