I don't suppose there's a highly-accessible curated database of hypotheses which appear to have tested very differently between mice (or other subjects) and humans.

You may find some of the reviews and meta-analyses in http://www.gwern.net/DNB%20FAQ#fn97 to be of interest!

I'm not sure about the FDA, though. IIRC, they only require all human trials to be reported, and they don't release the information. (This is why the recent Tamiflu meta-analysis was such big news: because the researchers managed to force the disgorgement of all studies' data and show poor performance of Tamiflu & systematic bias in which studies got officially published.)