Hierarchical latent class models for cluster analysis
Eighteenth national conference on Artificial intelligence
Gene Selection for Multi-Class Prediction of Microarray Data
CSB '03 Proceedings of the IEEE Computer Society Conference on Bioinformatics
| Hi-index | 0.00 |
Cluster methods have been successfully applied in geneexpression data analysis to address tumor classification. Bygrouping tissue samples into homogeneous subsets, moresystematic characterization can be developed and new subtypesof tumors be discovered. Central to cluster analysis isthe notion of similarity between the individual samples. Inthis paper, we propose latent structure models as a frameworkwhere dependence among genes and thus relationshipbetween samples can be modelled in a better way in termsof topology and flexibility. A latent structure model is aBayesian network where the network structure contains atleast a rooted tree including all variables, only variables atthe leaf nodes are observed, and the structure after deletingall the observed variables is a rooted tree. The maingain in using latent structure models is that they provide aprincipled and systematic method to handle the dependenceamong genes. There are other benefits offered by latentstructure models. They do not require any prior knowledgeon the determination of tumor classes and choice of similaritymetric, which are two important issues associated withthe traditional clustering techniques. They are also computationallyattractive due to the simplicity of their structures.We develop a search-based algorithm for learninglatent structures model from microarrays. The effectivenessof the algorithm and the proposed models is demonstratedon publicly available microarray data.