Gene expression data analysis of different brain areas based on non-negative matrix factorization

It is accounted that various regulatory activities between genes contain in the gene expression datasets. Alzheimer's disease (AD) are characterized by its hidden onset, complex pathological mechanism, hard diagnosis and it is difficult to reconstruct the genes signal pathways and its regulatory network. In this work, we improved nonsmooth nonnegative matrix factorization (nsNMF) to identify significant genes of Alzheimer's disease (AD) using cophenetic correlation coefficient to confirm the clustering number k. Since gene expression dataset has high noise, and the underlying information is hard to analyze according to the function of brain areas, we applied nsNMF to AD samples of hippocampus (HIP), entorhinal cortex (EC), media temporal gyrus (MTG) and primary visual cortex (VCX) which have close relationship of human learning and memory. After that, 3800 of significant genes were extracted including 10 known pathogenic genes. By biological analysis, many AD related biological process like apoptosis, metabolize and inflammation were obtained, and it is demonstrated that the improved nsNMF and the conjoint analysis method can deeply explore the pathways and gene regulatory ways of AD.