Friday, October 13, 2017

You maniacs! You blew it all up divided by zero!

The collective lidless eye of Pubpeer was recently directed at a vaccines-make-mice-autistic paper at the Journal of Inorganic Biochemistry... a paper now on the express train to Retractionville. For details, see "Mirror neurons and little men in boats" (a narrativised, hand-curated artisanal compilation of Pubpeer comments); or Orac's exhaustively-detailed account; or even the summary at RetractionWatch if you swing that way, we do not judge.

Much of that vast and unsympathetic attention was directed at Figures 2 and 4, where individual lanes from PT-PCR gels turn up repeatedly like a stamp collection with déjà vu, variously manipulated and freed from context, indicating the expression of several different genes in an admirable display of parsimony.

But the allure of those figures may have distracted attention away from Figure 1... unfairly so, for the validity of Figure 1 is also crucial for two previous papers by the same authors, in OA Autism and Immunotherapy, by dint of its pre-publication there.

Some of the initial commentary perseverated on the fifth lane (control mouse #3) in the TNF gel-slice of Figure 1C, for this is as blank as the Bellman's map: increased contrast reveals a neat rectangle of blankness, and the Pubpeer contributors speculated whether this was a thumbprint of image-enhancement deletion of something (e.g. with Photoshop), or an innocent artefact of high-loss low-resolution JPG compression of an image that was naturally blank. Only the original data can resolve the quandary. But let me explain.

The authors had made semi-quantitative measurements of the RNA expression for 18 proteins in the brains of their lab animals, and also the actual protein levels, where one protein was Actin or ACT as an internal-control test of reliability (for ACT is a housekeeping gene, always turned on). Of the 14 male mouse brains available -- seven exposed to aluminium, and seven controls -- three of each were chosen, by unspecified criteria. The outcomes for each RNA (and each protein) were three data points -- for the six samples were paired up at random, and converted to three Al:control ratios, as if each pair were After-and-Before measurements on the same mouse. I am not making this up. I had to read the explanation several times to reassure myself that the procedure was not the figment of my over-heated imagination. Any of the other five possible pairings would have yielded different ratios.

In Figure 1B, the three data points per RNA are expressed as a mean and standard error, as if they were a real distribution, purportedly differing significantly from zero in seven cases (according to one-sample t-tests). Why not plot the three ratios themselves? But wait, the engarbagement gets better!

Panels 1C and 1D document the same process of extracting expression ratios for arbitrary pairs, for seven actual proteins -- the ones netted in the fishing expedition of 1B. This is presented as independent verification, as if protein levels are separate evidence from the RNA that generated them. In the event, the replications are impressively similar: for example, the mean expression ratio (fold change) for TNF-A is 3 ± 1 in 1B (with p < 0.05) and again in 1D (with p < 0.01). What are the odds?

This is especially baffling because 3rd-control-mouse TNF-A is zero for one of those pairs (that blank rectangle in Panel 1C); THEY DIVIDED BY ZERO so the fold change for that pair → .

Then they calculated mean = 3 and standard error = 1 for that distribution of three points, one of which was INFINITY, that's not how maths works. So either that lane was naturally blank (and only a rectangular hole because JPG), and the stats are bogus; or it initially contained a non-zero protein trace that Photoshop taketh away, and Figure 1C is bogus [h/t Mitracarpus Capitatus]. Neither option should gladden the hearts of the editors of Immunotherapy or OA Autism.

The initial response of authors Shaw and Tomljenovic was to accept the gravity of the problem, accepting the need for retraction in a passive-voiced manner that blamed the absence of the homework upon the appetites of the dog. "Some images have been altered." "Data had been compromised." Original data had left the lab.

Ah, Passive Voice: Is there anything that can't be accomplished by it?

It is not clear whether they accept that Figure 1 is problematic, or view the alterations as confined to Figures 2 and 4; nor when the alterations occurred. Before or after Shaw submitted the 2017 paper to J.Inorg.Biochem? Were they present in the Powerpoint version of the study -- "Gene-toxin synergy in the brain of autistic mouse model” -- presented in 2015 to the 11th Keele Meeting on Aluminium? Anyway, in a subsequent interview the authors downplayed the importance of those fabricated Figures, as if retracting the retraction:
As for Shaw, he says the altered images “were not significant anyway.”
Enquiring minds have noted that Shaw and Tomljenovic have a chapter in a new book, due in the bookshops in two weeks' time, on "Neurodevelopmental toxicity associated with the use of aluminum vaccine adjuvants" and covering similar ground to the retracting-in-the-cold 2017 paper -- for Elsevier taketh away and Elsevier giveth:

Is it irresponsible to speculate that the altered figures are necessarily unimportant because they feature in the forthcoming chapter, and it is too late to change them? It would be irresponsible not to speculate.

Perhaps the editors of the book should be appraised of these concerns, no wait:
Edited by Christopher A. Shaw, Edited by Claire Dwoskin, Edited by Lluis Lujan , Edited by Lucija Tomljenovic
[Thx Rosewind]
The Table of Contents is a veritable Who's Who of the Aluminati (where "who's who" is a failed anagram of "clown car"). As well as Chapter 10 from Christopher Exley, there are contributions from both Gherardi's and Shoenfeld's groups; and Chapter 24 is Dr Little re-telling her "Sluts who use Gardasil are Punished with Infertility" campfire tale from five years ago.

But the book is not all Aluminium, for Brian Hooker is there with another version of his Alt-Stats MMR-causes-autism stylings (evidently Andrew Wakefield was not available, so instead, Chapter 27 is from Wakefield supporter and "retired sewage sludge researcher" David Lewis, with another attempt to defend medical malpractice and fraud). There is a chapter from Judy Mikovits, famed for proving that a laboratory contaminant causes Chronic Fatigue, who exchanged a career in science for one as Brave Maverick Outsider who Speaks Truth to Power Whatever the Consequences.

I am looking forward to the reviews.

No comments: