3PNMF-MKL: A non-negative matrix factorization-based multiple kernel learning method for multi-modal data integration...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biomedical big data integration and gene signature detection are complex challenges.
• Multi-modal data analysis requires sophisticated feature mining techniques.

Purpose of the Study:

• To propose a novel framework, 3PNMF-MKL, for multi-modal data integration and gene signature detection.
• To enhance the accuracy and efficiency of gene signature discovery from complex biological datasets.

Main Methods:

• Applied limma for feature extraction from individual molecular profiles.
• Utilized three-factor penalized non-negative matrix factorization (3PNMF) for data fusion.
• Employed multiple kernel learning (MKL) with soft margin hinge loss for classification and AUC estimation.
• Identified gene modules using average linkage clustering and dynamic tree cut.

Main Results:

• Developed a 50-gene signature from The Cancer Genome Atlas (TCGA) acute myeloid leukemia dataset.
• Achieved a high classification Area Under the Curve (AUC) score of 0.827.
• Demonstrated superior performance compared to state-of-the-art methods in AUC computation.

Conclusions:

• The proposed 3PNMF-MKL framework effectively integrates multi-modal data for gene signature discovery.
• The identified gene signature holds potential for cancer classification and understanding.
• The algorithm is adaptable for various multi-modal datasets in biological research.