Predicting gene phenotype by multi-label multi-class model based on essential functional features.

Area of Science:

• Genetics and Bioinformatics
• Computational Biology

Background:

• Phenotype prediction is crucial but complex due to genotype-environment interactions.
• Current methods for obtaining phenotype-associated gene/protein data are costly and time-consuming.
• Existing functional network-based predictions face challenges with complex data structures.

Purpose of the Study:

• To improve the accuracy and efficiency of phenotype prediction for multiple phenotypes.
• To develop a novel computational tool for identifying gene-phenotype associations.

Main Methods:

• Extracted functional enrichment features from Gene Ontology (GO) and KEGG databases.
• Utilized node2vec for learning gene functional embedding features from gene-gene networks.
• Applied feature selection methods (Boruta, mRMR) and incremental feature selection with RAkEL and base classifiers for model building.

Main Results:

• Developed an optimal multi-label multi-class classification model for phenotype prediction.
• Identified numerous literature-supported gene-phenotype associations.
• Discovered potential new phenotypes for candidate genes, validated computationally.

Conclusions:

• The proposed bioinformatics approach enhances phenotype prediction accuracy and efficiency.
• This method offers a valuable computational tool for genetic research and phenotype discovery.
• The findings facilitate a deeper understanding of gene function and its relation to observable traits.