Multi-omics data fusion using adaptive GTO guided Non-negative matrix factorization for cancer subtype discovery

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Personalized cancer treatment relies on accurate subtype discovery.
• High-dimensional multi-omics data (genomic, transcriptomic, proteomic) present challenges for integrative analysis due to heterogeneity.
• Identifying latent structures and correlations within and across omics data is crucial for understanding cancer.

Purpose of the Study:

• To develop an effective integrative analysis framework for cancer subtype discovery.
• To encapsulate the heterogeneity of biological mechanisms within multi-omics data.
• To predict homogeneous subgroups of cancer patients for targeted therapies.

Main Methods:

• Improved sparse-joint non-negative matrix factorization (sparse-jNMF) was developed.
• Initialization of sparse-jNMF was enhanced using the adaptive gorilla troops optimizer (Ada-GTO) meta-heuristic algorithm.
• Consensus clustering was employed to generate a patient-similarity matrix for robust patient subgroup identification.

Main Results:

• The framework was applied to four real-world multi-omics cancer datasets (colon adenocarcinoma, breast-invasive carcinoma, kidney-renal clear-cell carcinoma, lung adenocarcinoma).
• The proposed method achieved superior silhouette scores and cluster purity compared to classical initialization and other meta-heuristics.
• Kaplan-Meier survival analysis revealed statistically significant differences (p < 0.05) between patient clusters, outperforming iCluster, and identified somatic mutations for targeted treatments.

Conclusions:

• The Ada-GTO guided sparse-jNMF framework offers a robust approach for cancer subtype discovery using multi-omics data.
• The proposed meta-guided framework demonstrates superior performance over existing state-of-the-art methods.
• This approach holds significant potential for identifying homogeneous subgroups in other complex diseases.