SoylentNews is people
SoylentNews
Two decades of Alzheimer's research may be based on deliberate fraud that has cost millions of lives
Over the last two decades, Alzheimer's drugs have been notable mostly for having a 99% failure rate in human trials. It's not unusual for drugs that are effective in vitro and in animal models to turn out to be less than successful when used in humans, but Alzheimer's has a record that makes the batting average in other areas look like Hall of Fame material.
And now we have a good idea of why. Because it looks like the original paper that established the amyloid plaque model as the foundation of Alzheimer's research over the last 16 years might not just be wrong, but a deliberate fraud.
The suspicion that something was more than a little wrong with the model that is getting almost all Alzheimer's research funding ($1.6 billion in the last year alone) began with a fight over the drug Simufilam. The drug was being pushed into trials by its manufacturer, Cassava Sciences, but a group of scientists who reviewed the drug maker's claims about Simufilam believed that it was exaggerating the potential [...] and hired an investigator to provide some support for this position.
[...] In 2006, Nature published a paper titled "A specific amyloid-β protein assembly in the brain impairs memory." Using a series of studies in mice, the paper concluded that "memory deficits in middle-aged mice" were directed caused by accumulations of a soluble substance called "Aβ*56." [...]
That 2006 paper was primarily authored by neuroscience professor Sylvain Lesné and given more weight by the name of well-respected neuroscientist Karen Ashe, both from the robust neuroscience research team at the University of Minnesota. [...]
The results of the study seemed to demonstrate the amyloids-to-Alzheimer's pipeline with a clarity that even the most casual reader could understand, and it became one of—if not the most—influential papers in all of Alzheimer's research.[...]
What intrigued Schrag when he came back to this seminal work were the images. Images in the paper that were supposed to show the relationship between memory issues and the presence of Aβ*56 appeared to have been altered. Some of them appeared to have been pieced together from multiple images. [...]
Now Science has concluded its own six-month review, during which it consulted with image experts. What they found seems to confirm Schrag's suspicions.
They concurred with his overall conclusions, which cast doubt on hundreds of images, including more than 70 in Lesné's papers. Some look like "shockingly blatant" examples of image tampering, says Donna Wilcock, an Alzheimer's expert at the University of Kentucky.
[...] And it seems highly likely that for the last 16 years, most research on Alzheimer's and most new drugs entering trials have been based on a paper that, at best, modified the results of its findings to make them appear more conclusive, and at worst is an outright fraud.
Some interesting stuff between the [...] was cut down for this summary, so I recommend reading the linked story. I also coincidentally just listened to the most recent Science podcast where they go into this in much greater detail and is well worth a listen. [hubie]
(Score: 4, Insightful) by quietus on Sunday July 24 2022, @09:48AM (18 children)
Think statistics for a moment, people. The number of research papers published each year is well over a million. Number of papers offered for review, but not making it to publication, is a multiple of that number.
Correctness of these papers, like all natural phenomena, follows a normal distribution aka a Gauss curve. That means you’ve got at the low end about 5-7% duds through sheer incompetence, and at the high end an equal percentage of duds through outright fraud, even — or especially — after having passed through human review.
It is hence inevitable that some “key paper” turns out to be a dud either way. That does not mean, however, that the large majority of papers are frauds, attempts to put the little man down, or attempts to suck the government’s tit(s).
(Score: 2) by c0lo on Sunday July 24 2022, @10:21AM (11 children)
Assuming a Gauss distribution is quite an assertion.
Boltzmann distribution [wikipedia.org] is still a thing and explains quite well why 90% of everything is crap [wikipedia.org] - any explanation you can set forth on why scientific articles would make an exception?
https://www.youtube.com/watch?v=aoFiw2jMy-0
(Score: 3, Touché) by quietus on Sunday July 24 2022, @10:53AM (8 children)
If 90% of everything is crap, wouldn’t that imply that 90% of all buildings collapse under their own weight?
Your Boltzmann distribution is about a closed system with only 2 variables. Gauss curves on the other hand are observational and representative of natural phenomena e.g. take a random sample of a 1000 people and put out a frequency distribution of weight, length, skin color or whatever biological property, and you’ll get a normal distribution.
I prefer thousands of these observations over whatever good sounding quip of a single human (and writers often seem to me to be heavily overrated btw: like trusting a news anchor with your investment decisions).
(Score: 0) by Anonymous Coward on Sunday July 24 2022, @02:09PM (4 children)
We're talking about buildings, not piles of rubble, right?
What do the fact that more than 90% of the buildings no longer exist after 100 years, and yet there a enough of them that are valuable enough to be preserved even if they can't stand on their own (like the pyramids or Stonehenge do)?
Oh really? So, that energy can take zillions of forms doesn't matter to you, eh?
And this is in relation with the quality of the scientific publications exactly how?
Are you trying to say that all humans naturally exhibit the metabolic function of producing scientific articles? Or that "weight, length, skin color or whatever biological property" are somehow correlated with the quality of the scientific articles they produce, so we can soundly extrapolate based on this correlation?
You have thousands of observations that in regards with the quality of the scientific articles? Then it shouldn't be too hard to show those numbers.
If you don't, your assertion "is a Gauss distribution" have no basis so far.
(Score: 2) by quietus on Sunday July 24 2022, @02:57PM (3 children)
Enlighten me to the zillions of forms energy can take.
Also: are you saying that the human brain is not biological?
(Score: 2) by c0lo on Monday July 25 2022, @12:27AM (1 child)
Some examples for you, just don't expect to list them all
I can tell you've been nurtured on some degree and your behavior isn't entirely governed by your nature, but I reckon that nurture didn't go far enough. I can say this based on the fact that at least you can write (so you had some nurture) your naive and horrible simplified models of reality on S/N and your propensity of thinking they are adequate.
The bliss of your ignorance likely feels good to you. And, from this "perspective, people" PoV, the same likely happens to the entire cohort of people on this Earth that think their ignorance is as good as the scientists' knowledge.
---
As for the personal reason I chose to debate this very assertion of "Gauss distribution being applicable for professional performance": it reminds me the PHBes of this world using it in rank and yank [wikipedia.org]. It's even more reprehensible to me to see someone outside the area of scientific research using it as the gospel to characterize the quality of scientific papers: it may be so but, if you can't show me the numbers, have some decency and STFU before you draw conclusions based on not verified/demonstrated assumptions
https://www.youtube.com/watch?v=aoFiw2jMy-0
(Score: 2) by quietus on Monday July 25 2022, @11:53AM
Oopsie -- there I thought that there were only really 2 forms of enery: potential energy and kinetic energy. I guess I must stand corrected here.
As for the compliments: accepted with gratitude.
(Score: 3, Informative) by PiMuNu on Monday July 25 2022, @06:34PM
To be clear, Boltzmann distribution indeed occurs in nature; it describes the probability distribution of kinetic energy of atoms in a gaseous medium having temperature T.
Boltzmann distribution arises as a probability distribution based on the number of possible states of a gaseous medium. It's not about "forms of energy". I don't think "forms of energy" means anything.
(Score: 0) by Anonymous Coward on Sunday July 24 2022, @05:53PM
Survivorship bias? The buildings you see are the one out of ten that stand. :)
Besides, I wouldn't claim 10% success rate, but his argument does have an example for 25%:
(Score: 2) by hendrikboom on Sunday July 24 2022, @09:34PM
The survival of medieval cathedrals during their construction seems to follow statistics like that. Quite a few fell down before they were finished and had to be redesigned and rebuilt.
(Score: 1) by khallow on Thursday July 28 2022, @11:33AM
You won't get normal distributions, let us note. Human height, for example, has a sexual dimorphism that isn't covered by a normal distribution. Further, just consider biological gender itself. That basically consists of two large slots and a few weird side cases that are much less frequent. It's not even a continuous parameter that can be modeled by a normal distribution. You're deep in not-even-wrong territory here.
(Score: 3, Interesting) by hendrikboom on Sunday July 24 2022, @09:30PM (1 child)
I wonder what the energy and temperature of the correctness of a paper are.
(Score: 2) by c0lo on Monday July 25 2022, @01:08AM
Substitute the correctness for energy and the amount of grant money for the temperature and it may make sense? (large grin)
Look, I didn't say it is Boltzmann, I'm saying that Gauss is likely a wrong choice as it doesn't explain the Sturgeon's law - the later imply a skewness of the distribution that is not visible on very fine symmetrical shape of the Gauss bell.
Maybe it's a Pareto distribution [wikipedia.org]. Or maybe Poisson distribution is adequate to describe the chances of the number of crap papers being published any given year (I have my hunches it is not).
My point was why should I trust the poster that:
https://www.youtube.com/watch?v=aoFiw2jMy-0
(Score: 2) by PiMuNu on Sunday July 24 2022, @05:53PM
No need to guess. It has been well studied
https://en.wikipedia.org/wiki/Replication_crisis [wikipedia.org]
> In psychology ... 97 of the original studies had significant effects, but out those 97, only 36% of the replications yielded significant findings (p value below 0.05).
> In medicine ... Out of 49 medical studies from 1990 to 2003 with more than 1000 citations, 92% found that the studied therapies were effective.... 16 % were contradicted ... 44 % were replicated ... [the rest unclear]
> In medicine ... only 11% of 53 pre-clinical cancer studies had replications that could confirm conclusions from the original studies
Note that I believe the expected rate of disagreement between two having p-value 0.05, one passing and one failing, is given by the union of (possibility that the original test was a false positive) OR (possibility that the check was a false negative), which should be about 10 % of results do not get confirmed (to be precise 1-0.95^2). The wikipedia article doesn't go into details about the stats.
In medicine, a p-value of 0.05 seems to be requirement for a "discovery".
(Score: 1) by khallow on Tuesday July 26 2022, @05:51PM (4 children)
Unless you have policies in place to aggressively increase the cost of such (and weaker stuff like p-hacking) you'll get substantially increased representation of fraud in your high impact papers.
In particular there's no symmetry of a normal distribution here. Papers so low quality that they barely get published tell you nothing about the amount of fraud at the other end.
(Score: 2) by quietus on Wednesday July 27 2022, @12:45PM (3 children)
Those pesky p-values you find in research papers are based on a normal distribution, khallow, as anyone with a scientific or engineering background knows.
Brownie points for creativity in trying to shoe-horn statistical results in a Boltzmann curve, but I kindly suggest you return to that call-girl's belly button and whatever cocaine remains there.
(Score: 1) by khallow on Wednesday July 27 2022, @02:01PM (2 children)
(Score: 2) by quietus on Wednesday July 27 2022, @02:22PM (1 child)
The earth is flat and really supported by a pile of tortoises? There's a chain of pizza restaurants keeping kiddies in the basement for whatever?
(Score: 1) by khallow on Thursday July 28 2022, @11:30AM
The problem with the original supposition is that it is wrong. For a glaring example, no strictly positive valued natural phenomena are normally distributed by definition. A normal distribution infinitely extends into negative territory. Second, you can easily break a normal distribution merely by adding another with a normal peak at a different point. For example, even if the strictly positive variable of male and female human height (both natural phenomena, right?) were normally distributed, you have the peaks at different points. Thus, the category of all human heights, both male and female, would not be normally distributed.