Gene Ontology-based Knowledge Discovery Through Fuzzy Cluster Analysis

摘要

In an earlier paper we developed an algorithm based on linear combinations of Order Statistics (LOS) operators that constructs dissimilarity measures on pairs of gene products described by sets of terms from the Gene Ontology. Examples using a data set named GPD194_(12.10.03)showed that LOS measures produced relational data with more (visually apparent) clusters than BLAST-based groupings of the same data. In this paper LOS representations of gene product data are presented to both crisp and fuzzy clustering algorithms. We demonstrate how fuzzy partition matrices generated by Non-Euclidean Relational Fuzzy c-Means (NERFCM) clustering led to the discovery that two gene products in GPD194_(12.10.03) might not be properly annotated in the ENSEMBLE database. Revisiting the database, we found that the gene products in question had been reannotated since the initial sampling date (12/10/2003). Our examples also illustrate the ability of NERFCM to extract clusters that seem apparent in visual displays of the LOS dissimilarity data. These examples demonstrate the potential of fuzzy clustering for automation of knowledge discovery in gene product databases.