Multi Experiment Analysis

Clustering gene expression patterns is useful even if the numbering of the experiments has no physical meaning (as opposed to temporal patterns). Using the CAST algorithm, data [1] for 1246 C. elegance genes, from 146 experiments, was analyzed. The data was in the form log(red/green) (representing the log-ratio of the two sample intensity values at the corresponding array feature), per experiment. Contrary to the first trial, where the similarity measure needed to reflect the temporal nature of the data, the order of experiments here, in the total set, has little or no importance. Therefore, we use a unique similarity measure here. Figure 12.9 summarizes the results. For temporal data it makes sense to use other similarity measures when the corresponding sub matrices are clustered. Clustering the columns (rather than the rows) of the expression matrix is also possible and contains biologically meaningful information.

  

Figure 12.9: Top: Examples of the clusters found in analyzing the data for 1246 C. elegance genes. Below, cluster No. 2 has been enlarged. With a more precise version of this image, it is possible to identify regions that significantly contribute to correlations within a cluster, and then analyze the corresponding sub matrix. At the bottom, three examples are given, presenting clustering results for different genes in the same pre-defined family. Genes coding sperm proteins (8 genes, presented in A) are all clearly grouped together. The same is true for dehydrogenase genes (3 genes, presented in B). ATP related genes don't specifically correlate with any other pattern. This is expected since ATP is involved in many cell processes and is not correlated with specific conditions.