GC[Formula: see text]NMF: A Novel Matrix Factorization Framework for Gene-Phenotype Association Prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene-phenotype associations are crucial for understanding inherited diseases and drug development.
• Existing prediction methods are limited by sparse data and lack of integrated analysis of multiple factors.

Purpose of the Study:

• To propose a novel method, Weighted Graph Constraint and Group Centric Non-negative Matrix Factorization (GCλNMF), for improved gene-phenotype association prediction.
• To enhance prediction accuracy and enable interpretable representation of genes and phenotypes.

Main Methods:

• Developed GCλNMF by incorporating weighted graph constraints from hierarchical phenotype data.
• Utilized group prior information (intra-group correlation) among genes as a group constraint.
• Leveraged Non-negative Matrix Factorization (NMF) principles.

Main Results:

• GCλNMF achieved superior prediction accuracy on mouse and human gene-phenotype association datasets.
• The method demonstrated good understandability for biological explanations compared to state-of-the-art methods.
• Experimental results validated the effectiveness of integrating hierarchical phenotype and gene group information.

Conclusions:

• GCλNMF offers a powerful approach for gene-phenotype association prediction, advancing disease research.
• The method's interpretability facilitates biological analysis and understanding of complex gene-phenotype relationships.
• This work highlights the potential of integrating diverse biological data for enhanced predictive modeling.