A Novel High-Dimensional Kernel Joint Non-Negative Matrix Factorization With Multimodal Information for Lung Cancer...

Area of Science:

• Computational biology
• Bioinformatics
• Medical imaging analysis

Background:

• Correlating pathological imaging data with genomic information is crucial for clinical diagnosis.
• Identifying biological activities and biomarkers from tissue imaging features presents a significant challenge.

Purpose of the Study:

• To propose a high-dimensional kernel non-negative matrix factorization (NMF) method for fusing multi-modal information.
• To project RNA gene expression data and whole-slide images (WSI) into a common feature space for enhanced analysis.

Main Methods:

• Utilized multi-modal information fusion with a high-dimensional kernel NMF approach.
• Incorporated miRNA-mRNA and miRNA-lncRNA interaction networks as prior information.
• Employed radial basis kernel function for feature proportion calculation and orthogonal constraints for sparsity.

Main Results:

• The proposed NMF method demonstrated superior stability, decomposition accuracy, and robustness compared to traditional NMF.
• Identified lung cancer-related genes (e.g., COL7A1, CENPF, BIRC5) linked to tissue morphology.
• Discovered gene pairs with high correlation and potential survival biomarkers (e.g., CAPN8).

Conclusions:

• The developed NMF method shows potential for improving clinical diagnosis of lung cancer.
• The integration of imaging and genomic data facilitates the discovery of disease-related genes and biomarkers.