Self-righting or sinking?

Inspiration for this post comes from various sources, including Arjun Raj’s posts on the STAP papers (here and here) and that by The Spectroscope (here)
and my previous posts on the question of whether science does self-right.

I take issue with the trivialisation of data fabrication.

Duplication of lanes on a gel, which Arun Raj states might simply be due to sloppiness cannot be so – it has to be deliberate, if only because of simple geometry. The data are the image of the gel. So to duplicate a lane means to cut and paste. The practice of selecting the narrow band of interest for presentation is a horizontal selection on the image, whereas gel lanes run vertically. So cutting and pasting lanes is a very different operation and there is no reason to cut an paste vertically. The only reason I can think of is that the person is engaged in the business of making up data that is intended to be a simulacrum of the experiment that was never done.

So the problem with the STAP papers isn’t that they are wrong because all papers are wrong, in the sense that they are approximations representative of the understanding of their time or because the technique is tricky. They are wrong because data have been invented. The conclusions of the papers may be correct, the method may work in someone’s hands, but the data do not support the conclusions. If I had a penny for every great idea generated down the pub, I would own the universe. Making that evidence up cannot deemed to constitute evidence for a phenomenon any more than ideas sketched out on a soggy beer mat.

The physics mistakes alluded to in Arun Raj’s second post were just that, as far as I understand – I don’t think there was any evidence for fabrication, just the usual confounding of signal and noise. One might reasonably take issue with the publicity surrounding the “discoveries”, but there is no evidence for fabrication. So these are wrong for the “right” reasons, even if the mechanism of dissemination is deemed deplorable.

The question of why Obokata has taken most of the flak is interesting. I would agree that this may reflect more on her gender than on who is to “blame” for the fabrication of data. This in itself is a good reason to give the institution getting a good going over, but checking all data and placing the blame on the underlings will solve nothing. What really needs to be done is to look at structures and ask the simple question: why have we got a research structure where there is actual power vested in those at the “top” and why are those at the top virtually all male?

This leads neatly to The Spectrscope’s post where a simple question is posed: why is Nature, which supposedly peer-reviewed the papers not the subject of an investigation? As the publishers of papers clearly flawed from the point of view of data fabrication, why is Obokata and her section of Riken under the spotlight, but not Nature?
The answer is obvious. Nature and NPG do not engage in research, they make money (a lot of it) by selling the work of scientists to scientists. They are answerable to no one and will happily take your cash and do this again and again and again – it fits the business model perfectly. Since they are answerable to no one, the only way forward is for scientists and their libraries to vote with their wallets. Pay up suckers.

Simulacra of experiments is something that occurs sufficiently regularly in science as to generate a reaction from the community. Pubpeer is one of the first stops for people who have concerns with data in a paper, because journals are happy to receive the publicity from research, but, as we see in the STAP case, are not answerable when they fail to perform the most basic tasks we would expect to be part of pre-publication peer review. The most recent case brought to my attention is a paper describing the use of TrisNTA nanoparticles to label hexahistidine tagged proteins in electron microscopy experiments. The problems have, according to one of the peers, been raised with the journal editors, as well as being put up on Pubpeer (here)

As a user and fan of TrisNTA I have read the paper and comments on Pubpeer. Although I am not an electron microscopist, I have to agree that the suspicions of the peers seem well placed. I would also add that in our experience labelling proteins quantitatively in this way seems only to work in vitro with purified protein and monovalently functionalised nanoparticles, followed by purification of the protein-nanoparticle conjugate. As a bit of weird lab lore, we have observed in some cases that a percentage of protein is refractory to labelling with nanoparticles monovalently functionalised with TrisNTA (with TrisNTA nanoparticles in good molar excess). If this unlabelled protein is re-purified it will then conjugate to the nanoparticles. If that wasn’t weird enough, this second conjugation is at pretty much the same efficiency as the first, so not 100%. If we go through the process a third time with the re-purified unlabelled protein from the second conjugation we end up with the same labelling efficiency. We are baffled, puzzled and so on, as people who work with proteins generally are, and for want of a satisfactory explanation and an experiment that will test this, we have called this the “protein siesta problem” and content ourselves with purifying the conjugate! So it would be unusual to be able to label all the proteins in a complex in situ as done in this paper.

Regardless of the fate of this particular paper, the bottom line is that TrisNTA remains one of the best ways to label proteins with nanoparticles, since it combines the power molecular biology (hexhistidine tags being coded for in DNA) with that of chemistry, but proteins being proteins, you have to do your work and cutting corners will lead you nowhere.

To finish on a more positive note, while we have a continuous stream of critical comments on Pubpeer, often relating to data duplication of some sort (a further example here, perhaps) good constructive comments are also appearing (example here). We may be (slowly) turning a corner, but still have a long way to go.