It's come up before, most recently
here. The consensus seems to be that the method of displaying the information is really cool, but their data is crappy. The literature on supplements, and biomedical issues in general, is complex. It's hard to use data display software as your only interface to the literature and expect to get a good result. I think there needs to be some expert intervention somewhere between the raw data and the display. Either that or smarter software. I think that machine analysis of the scientific literature will, in the long run, be a tremendous advance. I just don't think we're there yet.