预印本调查：221 篇论文中 230 例本不该出现的蛋白条带

Here’s one of those disturbing-but-not-necessarily-surprising reports about the state of the literature. In this case, the preprint linked is the result of a deliberate search for papers that are working with a list of twenty cell lines that have clear genetic deletions and/or mutations. Then a search was done for Western blot experiments in these papers, with particular attention to cases where the blots show the presence of proteins that should not have been present at all. This is not a comprehensive sieving of the literature (it’s a one-man job), but it does illustrate the size of the problem.

230 such examples were found in 221 papers. Which isn’t so great, is it? There are several ways that this can happen, and all of them are mistakes that are well-known to researchers in the field, or at least they damn well should be. You can misidentify the cell lines that you’re using, for one thing, which obviously is a good way to invalidate all your work and all your conclusions thereafter. You can have the right cell line but contaminated with other cells entirely, and that’s going to send everything into the trash can as well. And finally, the antibodies that you’re using for those Western blots can be less specific than you’d hoped, or perhaps the wrong antibody entirely (no matter what you paid for and what the label says). 

These sorts of impossible-blot incidents have shown up many a time, but this is a more systematic look at them than usual. The most common anomaly is the report of caspase-3 in MCF-7 cells (indeed, one of those last links is to such an example), and then there’s BAX in Jurkat cells and STAT3 in PC-3s. Some of these are readily explained by the use of antibodies that are known to be garbage, and were probably known to be garbage even when the published work was being performed. But the reasons behind many of the reports remains obscure. Of particular interest is a series of papers from the same lab that relentlessly report the presence of STAT3 in PC3 cells, even though it is well-established not to be present in them at all.

Now that last link is to a journal that I have definitely never heard of, and which I doubt that anyone here has ever heard of either, and to which I will probably never return. So you might wonder, how much harm does this stuff do? But that’s too low a standard, I think, because it certainly doesn’t do any good. And in these days when we are told that AI/ML systems are going to take all the published literature and make sense of it for us, you have to think that feeding garbage into these systems is suboptimal. As the author of this new preprint says: “. . .this report is not meant to discourage the publication of contradictory results: unexpected results should be investigated and discussed, however, blithely introducing contradictions into the literature is unlikely to advance science”. Amen to that.

注意转载授权 · 阅读原文