Clustering models with applications in gene expression profiles
Open AccessMany of the currently available clustering models group subjects based on a type of distance measure, such as Euclidean or Mahalanobis distances. In gene expression data, genes in the same functional group tend to have expression levels that have higher correlations than genes in different groups. We propose an novel clustering model, the block-structured covaraince model (BCM) which assumes homogeneous correlations within blocks of the correlation matrix formed by group memberships. A maximum profile likelihood procedure motivated by an analogue under the stochastic block model, is proposed to estimate the latent group labels in BCM. Moment estimators are employed rather than maximum likelihood estimators when profiling out parameters. The exponential decay bounds and the uniform consistency of the moment estimators are proved under true or arbitrary sets of labels. An extensive simulation study showed that the BCM performs better than K-means and spectral clustering in .The second aspect of our research combines the normal hierarchical model with the SBM, through an integrative clustering method, so that the clustering information for both gene expression data and binary gene network are jointly modelled to estimate common underlying cluster structure with higher accuracy. It is beneficial when one or both of the data source are not reliable, which is especially true with noisy gene expression data. An empirical guideline in the choice between integrative clustering analysis versus separate clustering models is proposed. After applying the integrative clustering method to the mouse embryo single cell RNAseq and bulk cell microarray data, we find the integrative clustering model is capable of identifying not only the gene sets shared by both data sources but also the gene sets unique in each data source.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.