The blogger who spotted problems in Dana-Farber cancer papers | Hacker News Reader

The blogger who spotted problems in Dana-Farber cancer papers

nature.com

147 pointsrntn53 comments

Open articleView on HN

See also the 2021 New Yorker piece about Stanford researcher and doctored image spotter Elisabeth Bik.

"How a Sharp-Eyed Scientist Became Biology’s Image Detective"

https://www.newyorker.com/science/elements/how-a-sharp-eyed-...

For a few more example images/videos of the the image manipulations being caught, see here:

https://www.nytimes.com/interactive/2022/10/29/opinion/scien...

Or see her mastodon with photos: https://med-mastodon.com/@ElisabethBik

Some one should make where wordle of faked research images

Interesting to see how much of it is basically the Photoshop clone tool. I bet these are going to be harder to catch in the future, as fraudsters start using AI infilling instead. Or clone tool plus img2img.

One of my greatest professional worries is publishing a paper with an honest data analysis error and being accused of fabrication. For this reason, I spend an "unreasonable" amount of time triple-checking everything (my colleagues are annoyed by that).

I wish there was software that would help me spot mistakes by looking at my code and/or write up, but I also recognize that it would be abused by those committing fraud.

This is also one of my fears, and my strategy has been openness - I share my analysis/code with colleagues/Github/as supplement. Maybe they can spot errors or reuse some of it, and I feel that there is some "protection" in it

It should be a requirement that every paper be completely reproducible, for any software part of the paper. On any OS or systém, it should work. Similar to requiring statistics to be reported a certain way, there should be guarantees that the code should be available and work in all circumstances.

Unfortunately, I don't think this is realistic. Is it provable?

Does it require use of an "official" programming language and libraries, that are guaranteed to be maintained? Is Turbo Pascal for MS-DOS still supported? That's what I used for my dissertation, in 1993.

It's a far cry beyond what's expected in branches of science that are not mainly about programming.

Perhaps require making a docker container and writing a test script that includes inputs and expected outputs. There could be a website that loads those docker images and shows that the tests pass. Tests could be referenced within the paper. Would still require scrutiny to make sure that code doesn't hard-code the outputs.

This should be the way. Folks will forgive sloppy code on a “fast” paper. However it may invite more replication attempts.

Many folks have messed up simple algorithms before.

Are there (m)any examples of that? Especially given the fact that exposures of clear fraud when exposed is so often just ignored

I do the same. I write and prepare figures for each single paper that comes out from my lab. I double check each single Jupyter notebook and I share every single piece of raw data ( for the last paper we just submitted yesterday we uploaded 316Gb of data on zenodo). All this takes me weeks of course. I would be very skeptical of colleagues who publish more than 4-5 papers a year.

Why call Sholto David a "blogger" when he is properly a scientist?

Someone please give him a budget.

And also to Elisabeth Bik!

> The DFCI (...) had already been investigating some of the papers in question before David’s blogpost was published, says the centre’s research-integrity officer, Barrett Rollins. (...) (Rollins is a co-author of three of the papers that David flagged and is not involved in investigations into them, says DFCI spokesperson Ellen Berlin.)

How naive do they think people are?

Doubly so because comments on some had been out for a while. My guess is nothing was actually going to happen prior to the blog post.

It's not just Dana-Farber and that Alzheimer's lab. This type of fraud is endemic in biology. Occasionally it gets noticed. Behold, another identical thread on the same site about the scientists of the Breast Cancer Research Foundation from four years ago:

https://forbetterscience.com/2020/01/20/the-wizard-men-curin...

Featuring a scientist who was rumbled for image cloning, and then submitted still more Photoshopped images in his corrections (accepted by the journal, of course, because why wouldn't you let a detected fraudster have a second try with higher quality fraud).

First comment: "I was at Cornell….you are still missing at least my ex-Boss and collaborators."

It feels like there should be some kind of tool analogous to valgrind/ASAN/TSAN/UBSAN for code that is just run on all the images of any kind of paper before it is published to check for these kinds of things. Some of them just seem to be honest mistakes like linking the same image in two different places and this would fix that so nobody needs to bother checking for it in the future. It would also have the byproduct of catching all the malicious image doctoring that some people do which is just horrific for science in general.

Such tools exist, ex:

https://imagetwin.ai/

Wouldn't that just give the fakers a tool that lets them easily tweak their fakery until it can't be easily caught?

The saving grace here is that many of these scammers don't have the sophistication or expertise in the wide possibility space of fraud discovery techniques to know where they're exposing themselves. Every type of check that we make automatable and trivially repeatable by anyone will immediately cease to detect anything except the most lazy or incompetent scammers.

Publications should be responsible for the veracity of the content they publish. Instead they just shift the blame. They are 100% about profits and provide zero advantage, no wonder they are being replaced by free alternatives. But it doesn't solve the problem with fake papers and plagiarism.

> Publications should be responsible for the veracity of the content they publish

That is ridiculous and betrays a serious misunderstanding of the role of publications. The publications are a forum for discussion. People say incorrect things in discussions all the time, and they can’t be corrected until they’ve been said. Where do you propose they get said? How do you propose the publication “know” what’s true or not? There’s no way to do it, otherwise why would you be publishing what you’re publishing? The whole point is that people don’t already know what you’re telling them.

Scientific publishing is super broken in many ways but this is a terrible solution.

And yet, they do try to reject papers for various issues that at least approximate veracity. No journal will accept a paper proposing a perpetual motion machine. What is that but a veracity check? What is peer review? Glorified spellcheck? No, we expect publishers to at least catch the obvious stuff.

If you design a publisher around the idea that it truly has no responsibility for veracity, then you get... Arxiv. In that scenario traditional publishers truly provide no value.

Yes it is glorified spellcheck. They’re checking your methods. If you claim to have built a perpetual motion machine they almost certainly will not reject your paper on the basis that “perpetual motion machines are impossible,” but on the basis that your methodology [almost certainly] has an error.

"Your methodology probably has an error [because you got the wrong answer]" is not meaningfully different from "we're rejecting your paper because it's false". My point stands.

Ed: your other comment about how publishers used to just be mailing lists is actually kind of funny in how it proves my point. If we wanted publishing to just be a mailing list... we can just have a mailing list. Or use Arxiv. But today's publishers have to at least pretend to do better than that, which they mostly do by supposedly filtering papers on quality, which in turn is, you guessed it, 80% veracity checks (oh, and also increasing the right people's citation counts).

Well yeah when you add stuff I didn’t say, it does sound ridiculous.

I didn’t say they would reject it based on the result and presuming an error. They would find and reject the error (or on the basis of a million far more superficial issues which are actual good targets for reforming scientific publishing).

What did I add that's meaningfully different from what you said, i.e. in terms of outcome?

Ed: ok, I think I get it, you mean they'll somehow identify a specific method error, and if they can't then they'll... Just accept it? I simply don't believe that. I think the journal both should and actually will reject papers that violate the second law of thermodynamics without finding a specific methodological flaw.

If the paper is completely methodologically sound and the only problem is that it violates the second law of thermodynamics, then we need to take another look at the second law of thermodynamics

You're assuming that (a) reviewers not finding any methodology flaws in the paper means there are none, and (b) no flaws in the paper means no flaws in the work. Neither of those are good assumptions. And that's just the generically applicable arguments.

For the Second Law, Arthur Eddington had this to say:

> ...If someone points out to you that your pet theory of the universe is in disagreement with Maxwell's equations - then so much the worse for Maxwell's equations... But if your theory is found to be against the Second Law of Thermodynamics I can give you no hope; there is nothing for it but to collapse in deepest humiliation.

https://www.goodreads.com/quotes/947685-the-law-that-entropy...

> You're assuming that (a) reviewers not finding any methodology flaws in the paper means there are none, and (b) no flaws in the paper means no flaws in the work

No, you are assuming that with your flawed understanding of what a journal is and what peer review does. Like this isn’t a matter of opinion. Peer review in fact and by design is not a stamp of veracity. The reason is exactly the conundrum you’ve backed into.

If someone produces apparently high-quality science that challenges a well-regarded theory, it is only under your model that paper cannot be published. Under our model (the one currently in use today), the reviewers are expected to try to find holes in the methodology and, if they can’t find them, publish the paper anyway even if they disbelieve the conclusion. That way the broader scientific community can attempt to blow holes in the paper, and they frequently find things the authors and the peer reviewers missed. That is not a shot against the publisher: that is how science must work because of exactly the dynamic you’ve identified.

You're still not paying attention to what I'm saying, which is that journals are expected to catch the easy errors, not make an absolute determination of truth. I said this in my first comment on this thread.

Remarkable you can have this opinion about what should happen and then one comment later run into the exact reason that this is a terrible idea.

Failing to find an error doesn’t mean the paper is true. Correct! That’s why passing peer review isn’t and cannot be a badge of truthfulness.

That seems more like what blogs and posts are for, not academic journals. Unfortunately things have degraded so much that there is very little difference between a blog and a research article, but there should be a very large distinction.

No, this is what journals are and always have been. Historically they were much looser even. Essentially just mailing lists of interested parties.

Odd to call a molecular biologist a 'blogger' in the title when the reality is they're highly trained experts. Really shows the lengths some academics go to falsify data.

How does funding this work work? Is it donation? Are there bug bounty programs for research out there? (I quitely wonder if this would not add a lot of credibility to research institutions again. I just wonder if we as state funded institution would be allowed to pay this)

The blog does ask for donations at the end of the article, so it is at least partially donation-funded

As another commenter here put it, this is "endemic in biology," unfortunately.

I once had a job at a top research lab doing engineering on their HPC clusters. There are a lot of big-ego fraudsters in science.

I think the most egregious thing I knew about was that payments, I mean donations, would come in from the nearby Roman Catholic Diocese to certain researchers. A few months later, something like a research paper supporting the idea that genital mutilation (male circumcision) reduces AIDs transmission rates would come out.

Things like faking, I mean "correcting" datasets and just making stuff up wasn't uncommon at all. You had to publish or perish.

> I used to write letters to the editor [at journals], but it’s a very infuriating process.

This is a shame

>design some file-organizing system that involves giving research images a sensible name. And then when it comes to checking your paper before you publish it, you need to trace back all the images to the raw data and check them against the metadata. For example, if you have a photo labelled as ‘Day three’, does that correspond to the date the photo was taken or that the experiment happened on?

Blockchain, but with human-readable metadata? Sure sounds like it is do-able.

No, a notepad.

Data is taken from other papers, it cannot be centralized.

Since we're talking about immutable data...sure it can. If you really need to prove provenance of the metadata, surely a checksum is enough?

You just photograph the notepad and release it as an artifact alongside the paper, if you don't feel like typing it up.

Or you skip the notepad and use a tablet and an electronic lab notebook. I know of a few researcher doing this successfully, but for this to work at scale senior scientists and group leaders need to force everyone in their labs to do the same

Absolutely a culture issue rather than a technical one!

Yes the technology was there already 15 years ago (minus a nicer UI and a slightly better UX). Researchers despite all the talk about innovation are surprisingly resistant to change

Or maybe just adopt the modern solution and use a electronic lab notebook (ELN). I'm a software engineer and we are trying to introduce ELN in our institution, it's not easy to convince researchers of the benefits of reproducibility. People really just use manual methods like "sensible names"

What is an ELN?

Electronic lab notebook.

An issue, and maybe you can speak to this from experience, is the dread of Enterprise Software.

The notebook has to support recording any kind of information, be accessible anywhere, even on tiny laptop screens, and otherwise stay out of the way. It should involve little or no manual data transfer. It should probably work harmoniously with git. For some of us, investing brain cells in the use of proprietary software might be a negative.

For my work, Jupyter notebooks are the best solution I've found. A wet chemist might have a different opinion.

There are plenty of non commercial, non enterprise ELN system. With all their down sides, they are all pretty similar and relatively easy to use (they could use better UI but that's often a problem in research). I think it's almost completely a cultural problem

No need for any blockchains here, this essentially describes an SDMS (scientific data management system). Which do exist, but are not very common outside of larger companies. And the more common Electronic Lab Notebooks (ELN) are in most cases only connected to the original data in a rather haphazard way, if at all.

You need to make the entire process digital, from measuring the data to tranferring and storing it and then linking it in your ELN or any other place you make use of it. That's a process that is only just beginning in many cases, if at all.

Image certification is already a thing https://rohitbhargava.com/leica-pioneers-the-anti-ai-camera-...

Reply on news.ycombinator.com