jNMFMA: a joint non-negative matrix factorization meta-analysis of transcriptomics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Omics data integration presents challenges due to heterogeneity.
• Existing methods for identifying differentially expressed genes (DEGs) often overlook data dependence structures, leading to high false positive rates.
• There is a need for advanced meta-analysis methods that account for underlying biological relationships in omics data.

Purpose of the Study:

• To propose a novel meta-analysis method, joint non-negative matrix factorization for meta-analysis (jNMFMA), for robust DEG identification.
• To leverage dependence structures within transcriptomics data for improved biological knowledge discovery.
• To develop a flexible method capable of integrating diverse omics data types.

Main Methods:

• Developed a joint non-negative matrix factorization (jNMF) approach to simultaneously decompose multiple omics data matrices.
• Mapped high-dimensional omics data into a low-dimensional space using metagenes representing hidden biological signals.
• Identified DEGs by associating them with differentially expressed metagenes derived from jNMF.

Main Results:

• The proposed jNMFMA method effectively identifies DEGs by utilizing underlying data dependence structures.
• jNMFMA demonstrates superior performance and robustness compared to existing popular approaches in both simulated and real-world cancer data.
• The method successfully extracts dependence structures, leading to more efficient and accurate DEG identification in meta-analysis.

Conclusions:

• jNMFMA offers a powerful and flexible approach for DEG identification in omics data meta-analysis.
• The method's ability to capture data dependence structures enhances accuracy and reduces false positives.
• jNMFMA provides a valuable tool for mining biological insights from heterogeneous omics datasets.