A Fuzzy Approach for Analyzing Outliers in Gene Expression Data

Outlier gene expression patterns identify abnormal gene behavior, possibly indicating the deviation in gene function for certain tumor types. It may also reveal novel gene-tumor relations, as well as novel tumor types. This is important in designing drugs for tumors as well as in studying the functional relations between genes. Apart from identifying outliers, associating outliers to the clusters in a gene expression dataset can reveal information about the source of outliers, i.e. to which tumors they are related. This work proposes a fuzzy approach which combines outlier detection and clustering results, to analyze gene expression outliers based on their relations to clusters. Both outlier detection and clustering are done based on a connectivity-based clustering algorithm, and the results are then combined using an iterative technique that propagates fuzzy memberships from patterns to their neighbors. Experimental results on leukaemia expression patterns are used to illustrate the proposed approach.