Résumé: Microarray technology is an important innovation that simultaneously facilitates measuring the expression level for thousands of genes in different samples. One basic and widely used technique in microarray data analysis is clustering. Due to some characteristics of microarray data related to noise, redundancy, and complex biological hidden process, a robust clustering method is required. In this paper, adaptive interval type2-possibilistic C-means and adaptive interval type2-possibilistic fuzzy C-means clustering methods are developed to better manage the mentioned characteristics and uncertainties. The proposed algorithm not only takes advantage of possibilistic C-means clustering and type2-fuzzy sets to handle noise and uncertainty, but also uses the concepts of "adaptive parameter" and "shadow sets" for each cluster per iteration. Therefore, based on the uncertainty of bandwidth, the permissive and strict bandwidths are defined, which makes the model more resistant to noise and outliers. In order to evaluate the proposed algorithms, synthetic datasets, UCI datasets, and microarray datasets of Alzheimer's disease (AD) have been implemented. Considering AD datasets, several clustering methods were initially evaluated through sample-based clustering, using comparative analysis. Then, the proposed methods were applied based on gene-based clustering, where heatmaps visualized the results. The outcomes of the experiments demonstrated better clustering performances for the proposed methods as compared to some well-known soft clustering methods.

Langue: Anglais