In a typical gene expression study with high throughput technique, such as microarray, a biologist usually focuses on the top genes ranked by the P-values to establish gene functional relationship / network, biological pathway, and microbiologically ramifications of the gene's selection. With more datasets publically available, researchers pool data from independent experiments, typically by pooling P-values with equal weight assigned to each dataset, aiming to fetch more biological information from the pooled data. However, the qualities of datasets may vary substantially. Assigning equal weights may not guarantee the optimal result. Applying the equal weights approach to six independent datasets, we observe the top rank genes of data pooled with this approach have less functional coherence than the single dataset that has highest functional coherence. We propose a procedure based on enhanced simulated annealing (ESA) and literature semantic indexing cohesive (LSI-c) analysis to assign optimal weights to datasets so as to maximize the functional coherence of the top rank genes ordered by their pooled P-values. We observe significantly more functional coherence in optimally pooled data than any single dataset or data pooled with equal weights. Identification of top rank genes through our optimal procedure should improve the downstream analysis.