VxInsight Validation: experimental dependence


Which gene clusters are highly dependent on the specific set of experiments used?


To address this question, we randomly divided the set of 553 experiments into two groups, and then formed two new gene expression terrain maps (the first random half and the second random half). We compared the first half, the second half and the full terrain maps to each other in two ways. In the first way, we coded each of the genes from one of the mountains from the full map with a different color, and then plotted the genes on each of the random halves. Many gene expression mountains remained clustered in all three terrain maps, indicating that these gene expression groups are formed by experiments present in both random halves. These gene clusters are probably not formed by a small number of specific experiments. Other gene expression mountains appeared differently in the two random halves, indicating that these gene clusters are formed by specific experiments present in one half but not the other.

In the second approach, we determined the nearest 200 neighbors for each gene in each of the random half-maps, and then calculated the number of genes that were in common. We plotted a histogram of the number of shared genes out of 200 neighbors. The graph shows that there are a large number of genes that share common neighbors when we would expect very few genes by random chance.

The data are shown here.