Gene prioritization using Bayesian matrix factorization with genomic and phenotypic side information.

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Traditional gene prioritization methods often analyze diseases individually, missing shared patterns across conditions.
• This limitation hinders comprehensive understanding and accurate prediction of gene-disease associations.

Purpose of the Study:

• To develop an advanced gene prioritization method that captures patterns across multiple diseases and phenotypes.
• To improve the accuracy of gene-phenotype matrix completion by incorporating diverse side information.

Main Methods:

• Formulated gene prioritization as a matrix factorization problem for a sparse gene-phenotype matrix.
• Extended classical Bayesian matrix factorization to integrate multiple side information sources.
• Developed a method to predict unknown gene-phenotype associations using available data.

Main Results:

• The proposed method demonstrated improved accuracy compared to the established Endeavour method.
• Successfully integrated gene and Human Phenotype Ontology (HPO) data for enhanced predictions.
• Showcased particular effectiveness in prioritizing genes for nervous system, eye, endocrine, metabolic, and congenital disorders.

Conclusions:

• The novel Bayesian data fusion approach offers a more effective strategy for gene prioritization than existing methods.
• This method facilitates more accurate identification of disease-associated genes, especially for complex or rare conditions.
• The developed package (Macau) provides a valuable tool for researchers in computational biology and genetics.