Figure 4

Execution time comparison of Parzen kernels superpositioning algorithm on one, two and three computers. The Parzen kernel is applied to a replicate series of microarrays to identify a reference conditional cumulative distribution to access statistical significance of observed differential expression. The number of possible comparisons in a replicate series with N genes and M arrays is NM-1, thus the computation can be quite time-consuming. The number of kernel evaluations is also dependent on the dimension of the resulting matrix, K, of the conditional distribution. The number of kernel evaluations would therefore be (K2.N) M-1. The speedup gained by using mGrid and separating the task into three independent subtasks for larger matrices (150 × 150 or 200 × 200) is 2.8. This is considered excellent as the maximum theoretical execution time speedup possible is 3. As the problem size increases, the data transfer time becomes relatively insignificant.