On the other hand, these are by definition high impact studies, ones Amgen wanted to try to develop drugs upon, and they wouldn't have gone public unless they were reasonably sure they'd given them a legitimate try. And I already assume at least 1/2 of biomedical research is junk ... and it wasn't just because I found organic chemistry easy and fun, I switched my major to chemistry in part because I didn't like the ... vibes, for lack of a better word, I got from the (MIT) biology department, aside from my adviser Phillip Sharp (who was clearly going to get a Nobel soon, and from solid research; Lord Baltimore, though ... did not impress).
My favorite professor, Jerry Lettvin, discovered a marvelous thing about frog's eyes: unlike/in addition to the basic edge detector sort of things we've got, they have a specific "bug detector", simulated in the lab with a bowl over the frog's head, and something black that's moved into the frog's vision via a magnet on the other side. And of course probe(s) in the optic nerve.
Another lab had difficultly reproducing this (this was seriously novel, and in theory easy and cheap to reproduce), so Jerry practiced for a while doing the procedure with side-cutters (https://en.wikipedia.org/wiki/Diagonal_pliers), and then showed up at that lab in his usual not clean shaven, simple and slightly dirty blue work shirt and black pants, and got the experiment to work with the lab's equipment (e.g. probe(s) and oscilloscope) ... and the side-cutters he brought ^_^.
(The frogs were, BTW, reported to apparently not be terribly hurt by this, and after healing up exhibited normal frog behavior. Then again, if I was in that lab, I'd be harvesting them for their tasty legs, as I did with all the bull frogs I shot with my BB gun growing up, in addition to dissecting them :-).
So "failure to reproduce on the first few attempts" does not mean "bad science".
In genomics, however, failure to reproduce was the norm for many years. There were a few spectacularly good early results that held up, but the ubiquitous use of cross-validation (which unless done with insane care is simply invalid) and analysis of significance that was frequently just wrong meant that a lot of results were published that were the numerical equivalents of early Royal Society papers on deformed cows and the like: meticulous descriptions of anomalous one-offs.
A lot of what is happening is generational: the older generation of biological researchers were never trained or equipped with anything like the analytical tools required to cope with the large numerical datasets that labs started generating in the '90's thanks to new technologies in the wake of the Human Genome Project.
http://www.reuters.com/article/2012/03/28/us-science-cancer-...
"A former researcher at Amgen Inc has found that many basic studies on cancer -- a high proportion of them from university labs -- are unreliable, with grim consequences for producing new medicines in the future.
During a decade as head of global cancer research at Amgen, C. Glenn Begley identified 53 "landmark" publications -- papers in top journals, from reputable labs -- for his team to reproduce. Begley sought to double-check the findings before trying to build on them for drug development. [...] Result: 47 of the 53 could not be replicated. "