Learning the local molecular pattern of Alzheimer's disease by non-negative matrix factorization

Gene microarray technology is an effective tool to monitor simultaneous activity of multiple cellular pathways from thousands of genes in a single chip. Many clustering methods have been developed to identify groups of genes or experimental conditions that exhibit similar expression patterns from gene expression data, such as hierarchical clustering, k-means, and self-organizing maps (SOM). The limitations of these clustering algorithms are: they group genes (or conditions) based on global similarities in their expression profiles and only assign each gene to a single cluster. In this work we present a biclustering method-nonnegtive matrix factorization (NMF) to avoid the above drawbacks and discover the local molecular pattern from gene expression datasets of Alzheimer's disease (AD). NMF can be applied to reduce the dimensionality of the data and describe the data as a positive linear combination of a reduced number of factors. By applying a sparseness enforcement variable into classical NMF, the more local structures with meaningful biological information inherent in the data are captured by clustering genes and samples simultaneously, and the classification of samples is well improved. The analysis and discussion of the identified local structures demonstrated that they related many pathways which play a prominent role in AD and the activation patterns to AD phenotypes.