Gene function prediction based on combining gene ontology hierarchy with multi-instance multi-label learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene function annotation is a critical challenge in the post-genome era.
• Vast amounts of genomic data require advanced methods for knowledge extraction.
• Machine learning offers a powerful strategy for mining genomic data and predicting gene function.

Purpose of the Study:

• To improve gene function annotation by integrating Gene Ontology (GO) hierarchy with a multi-instance learning framework.
• To develop an efficient computational method for predicting gene function using machine learning algorithms.

Main Methods:

• Enhanced multi-instance hierarchical clustering incorporating GO hierarchy.
• Application of multi-label support vector machine (MLSVM) and multi-label k-nearest neighbor (MLKNN) algorithms for gene function prediction.
• Validation using four yeast expression datasets.

Main Results:

• The proposed method effectively utilizes GO hierarchy within a multi-instance, multi-label learning structure.
• MLSVM and MLKNN algorithms demonstrated efficacy in predicting gene function based on the enhanced framework.
• Experimental results on yeast datasets confirmed the efficiency of the developed approach.

Conclusions:

• The integration of GO hierarchy with multi-instance learning significantly enhances gene function annotation.
• The developed machine learning approach provides an efficient and accurate tool for predicting gene function.
• This study contributes a valuable method for navigating complex genomic data in the post-genome era.