I am currently processing large scale RNA-seq data from independent cohorts.

Is there a systematic strategy to normalize the TPM of each transcript so that the expression levels can be compared without concerning the batch effect?

Thanks!

A critical question here is if you are interested in making comparisons within cohorts (e.g. comparing disease vs normal across numerous studies, each with normal and disease samples), or across cohorts (e.g. this study has disease A, this other study has disease B). The first kind of comparison is called "balanced" while the second one is "confounded".