What is the optimal balance provided by meta-analysis of p-values from individual dataset DE analyses in terms of type I error control and DE gene detection power?