Incidence of missing values in hierarchical clustering of microarrays data

Microarrays allow to determine expressed genes in a given cell type, at a given time and under particular experimental conditions. These experiments are performed on a huge scale and produce numerous data. Methods like Hierarchical Clustering (HC) [1] or Self Organizing Map [2] are often used to identify co-expressed genes. The data often contain several missing values (MV) due to experimental difficulties. These missing data represent a drawback for using certain quantitative analysis. Moreover, the presence of MV may strikingly modify the results. Usually, the users remove the genes with MV or replace the MV by a constant (zero or the average expression level of the entire experiment). Troyanskaya et al. demonstrate that the weighted K-Nearest Neighbours (KNN) seems to be the most accurate method to estimate the value of MV in microarrays data [3]. First we have analysed the quantity of missing values in different public sets of microarrays data concerning Saccharomyces cerevisiae. In a second step we have assessed the effects of MV and the chosen metrics in the HC results. Then we have tested the KNN method incidence on the HC with our different datasets.