Comment by mrtnmcc

You might also want the topic files to be compressed against each other to get a baseline matrix and then multiply any results by the inverse, assuming equal priors on the topics.