I am looking to set up a process to compare all datasets, whether it be RNASeq or microarray, for a specific cancer type in oder to look for subgroups based on different gene expression. There are plenty of cross-platform normalization methods, but are for singular experiments and normalize the data based on the other data present. As I would eventually like to expand to look for similar subgroups between cancers, I wouldn't be able to use these kinds of normalization processes as They would only be normalized for the initial datasets I used. I believe meta-analysis would be more appropriate here, so that I could use test statistics from datasets that have been normalized individually.

I have seen various types of meta-analysis for gene expression including using the Z-statistic, p-value, quantile based binning, etc. I was wondering whether 1. My line of reasoning is correct and 2. Whether there is a common best practice statistic to use to perform meta-analysis on datasets from different platforms?