Integrative network-based clustering of multi ‘omics data using non-negative matrix factorization

Oregon Convention Center 

Disease subtype discovery analysis using multi-source 'omics data in an integrative framework is a powerful approach. Such analyses leverage both between and within data correlations to identify latent subtype structure in the data. A new integrative similarity network-based clustering method is proposed using the non-negative matrix factorization, nNMF. The method utilizes the consensus matrices generated using the intNMF algorithm on each type of data as a network among the patient samples. The networks are then fused together to create a single comprehensive network structure optimizing the strengths of the relationships. A spectral clustering is then used on the final network data to determine the cluster groups. The method is illustrated with simulated, and real-life datasets obtained from The Cancer Genome Atlas studies on glioblastoma, lower grade glioma and head and neck cancer. nNMF works competitively with previous methods and sometimes better as compared to previous NMF or model-based methods. The novel nNMF method allows researchers to identify the latent subtype structure inherent in the data so that further association studies can be carried out.

Integration

nNMF

Latent

Network

Spectral 

Biometrics Section