AI Helps Unravel a Cause of Alzheimer's Disease and Identify a Therapeutic Candidate (ucsd.edu) 33
"A new study found that a gene recently recognized as a biomarker for Alzheimer's disease is actually a cause of it," announced the University of California, San Diego, "due to its previously unknown secondary function."
"Researchers at the University of California San Diego used artificial intelligence to help both unravel this mystery of Alzheimer's disease and discover a potential treatment that obstructs the gene's moonlighting role."
A team led by Sheng Zhong, a professor in the university's bioengineering department, had previously discovered a potential blood biomarker for early detection of Alzheimer's disease (called PHGDH). But now they've discovered a correlation: the more protein and RNA that it produces, the more advanced the disease. And after more research they ended up with "a therapeutic candidate with demonstrated efficacy that has the potential of being further developed into clinical tests..." That correlation has since been verified in multiple cohorts from different medical centers, according to Zhong... [T]he researchers established that PHGDH is indeed a causal gene to spontaneous Alzheimer's disease. In further support of that finding, the researchers determined — with the help of AI — that PHGDH plays a previously undiscovered role: it triggers a pathway that disrupts how cells in the brain turn genes on and off. And such a disturbance can cause issues, like the development of Alzheimer's disease....
With AI, they could visualize the three-dimensional structure of the PHGDH protein. Within that structure, they discovered that the protein has a substructure... Zhong said, "It really demanded modern AI to formulate the three-dimensional structure very precisely to make this discovery." After discovering the substructure, the team then demonstrated that with it, the protein can activate two critical target genes. That throws off the delicate balance, leading to several problems and eventually the early stages of Alzheimer's disease. In other words, PHGDH has a previously unknown role, independent of its enzymatic function, that through a novel pathway leads to spontaneous Alzheimer's disease...
Now that the researchers uncovered the mechanism, they wanted to figure out how to intervene and thus possibly identify a therapeutic candidate, which could help target the disease.... Given that PHGDH is such an important enzyme, there are past studies on its possible inhibitors. One small molecule, known as NCT-503, stood out to the researchers because it is not quite effective at impeding PHGDH's enzymatic activity (the production of serine), which they did not want to change. NCT-503 is also able to penetrate the blood-brain-barrier, which is a desirable characteristic. They turned to AI again for three-dimensional visualization and modeling. They found that NCT-503 can access that DNA-binding substructure of PHGDH, thanks to a binding pocket. With more testing, they saw that NCT-503 does indeed inhibit PHGDH's regulatory role.
When the researchers tested NCT-503 in two mouse models of Alzheimer's disease, they saw that it significantly alleviated Alzheimer's progression. The treated mice demonstrated substantial improvement in their memory and anxiety tests...
The next steps will be to optimize the compound and subject it to FDA IND-enabling studies.
The research team published their results on April 23 in the journal Cell.
"Researchers at the University of California San Diego used artificial intelligence to help both unravel this mystery of Alzheimer's disease and discover a potential treatment that obstructs the gene's moonlighting role."
A team led by Sheng Zhong, a professor in the university's bioengineering department, had previously discovered a potential blood biomarker for early detection of Alzheimer's disease (called PHGDH). But now they've discovered a correlation: the more protein and RNA that it produces, the more advanced the disease. And after more research they ended up with "a therapeutic candidate with demonstrated efficacy that has the potential of being further developed into clinical tests..." That correlation has since been verified in multiple cohorts from different medical centers, according to Zhong... [T]he researchers established that PHGDH is indeed a causal gene to spontaneous Alzheimer's disease. In further support of that finding, the researchers determined — with the help of AI — that PHGDH plays a previously undiscovered role: it triggers a pathway that disrupts how cells in the brain turn genes on and off. And such a disturbance can cause issues, like the development of Alzheimer's disease....
With AI, they could visualize the three-dimensional structure of the PHGDH protein. Within that structure, they discovered that the protein has a substructure... Zhong said, "It really demanded modern AI to formulate the three-dimensional structure very precisely to make this discovery." After discovering the substructure, the team then demonstrated that with it, the protein can activate two critical target genes. That throws off the delicate balance, leading to several problems and eventually the early stages of Alzheimer's disease. In other words, PHGDH has a previously unknown role, independent of its enzymatic function, that through a novel pathway leads to spontaneous Alzheimer's disease...
Now that the researchers uncovered the mechanism, they wanted to figure out how to intervene and thus possibly identify a therapeutic candidate, which could help target the disease.... Given that PHGDH is such an important enzyme, there are past studies on its possible inhibitors. One small molecule, known as NCT-503, stood out to the researchers because it is not quite effective at impeding PHGDH's enzymatic activity (the production of serine), which they did not want to change. NCT-503 is also able to penetrate the blood-brain-barrier, which is a desirable characteristic. They turned to AI again for three-dimensional visualization and modeling. They found that NCT-503 can access that DNA-binding substructure of PHGDH, thanks to a binding pocket. With more testing, they saw that NCT-503 does indeed inhibit PHGDH's regulatory role.
When the researchers tested NCT-503 in two mouse models of Alzheimer's disease, they saw that it significantly alleviated Alzheimer's progression. The treated mice demonstrated substantial improvement in their memory and anxiety tests...
The next steps will be to optimize the compound and subject it to FDA IND-enabling studies.
The research team published their results on April 23 in the journal Cell.
Those models are known (Score:4, Informative)
Re: Those models are known (Score:2)
What if instead of AI you used "the Attention mechanism" which is what made ChatGPT able to generate perfect grammar when all the previous AI machine learning techniques (neural networks, decision trees, markov models, what-have-you) failed?
Re: (Score:2)
Re: (Score:2)
Proteins are so a complex that machine learning model is the best approach
Definitely...since so many projects have done exactly that already. AlphaFold, ESMFold, Metagenomic Atlas, RoseTTAFold, etc. I'm pretty sure the technique is at the point these days that just about any protein structure of any interest can just be handed over to "AI" and it's as good as modeled.
Re:Those models are known (Score:5, Informative)
This is relatively new in the space. AlphaFold by DeepMind. The architecture is a variant of the transformer architecture that LLMs use.
This isn't the small statistical models that were used for this kind of stuff a decade and a half ago. It's fair to call this AI.
Sounds cool (Score:5, Interesting)
When they combine this with gene editing to introduce the Christchurch mutation ( https://www.alzforum.org/mutat... [alzforum.org] ) into many of the brain's cells, that's strong a path to cure.
Great news, I hope (Score:2)
As someone whose mother and grandmother both had serious Alzheimer's, I've got a personal interest. But in terms of the benefits to society of not having this major problem it's a great prospect. Let's hope this doesn't prove to be a false dawn.
Re: (Score:3)
My Mom too. You lose them day by day and year by year, and it's excruciating.
False starts are the norm in this kind of research, and this may well be another one, but that's how you get to success in science, by scaling a mountain of errors.
Hopefully this research won't be affected by cuts to NIH, but it probably will be.
Re: (Score:2)
My Mom too. You lose them day by day and year by year, and it's excruciating.
My dad passed from cancer and my mom from Alzheimer's. Dad was in the hospital for a couple weeks at the end. Mom was in long term care for 2 1/2 years before the end. No idea what my own fate will be, but given a choice I'd take the cancer over the Alzheimer's any day.
Remarkably complicated for a side effect (Score:3)
Is triggering gene acttion a side effect or is it an evolved process that actually does something important that they haven't discovered yet? It wouldn't be the first time that pulling on one metaphorical piece of bilogical string unwittingly unravelled something else.
Re: (Score:3)
Is triggering gene acttion a side effect or is it an evolved process that actually does something important that they haven't discovered yet? It wouldn't be the first time that pulling on one metaphorical piece of bilogical string unwittingly unravelled something else.
A good example is how some groups have evolved malaria resistance, and that made them vulnerable to sickle cell anemia.
So yeah, I've a strong suspicion that a gene that actually causes a major problem has some other effect that might be positive in earlier life.
Re: (Score:2)
Everything in biology is a remarkably complicated side effect. All aspects of biology are a horrendous mess and hey worse the deeper you look into them. Everything interacts work everything and anything can go wrong at any stage and there are probably yet more interacting mechanisms to deal work some of the wrongness.
Re: (Score:2)
Re: (Score:2)
He's saying something of which "engineering involves tradeoffs" is a subset. Biology is often(usually?) much messier than anything engineering would create, because the changes are made without intention, and random changes are filtered by "what survived".
E.g., no engineer would ever place the neural bodies between the light source and the photo-receptor. It's a random chance made by a primitive chordate, which was "good enough to work". (IIUC, octopus ancestors made the opposite choice, and that also wo
Funny and sad (Score:5, Insightful)
Re: (Score:2)
It's a buzzword for sure, but there is no denying that AI has, even before it became the latest buzzwords, been used in scientific research to great effect, distilling raw data into information that can be used for breakthroughs.
Re: (Score:3)
Alphafold (Score:5, Informative)
From the article published in Cell [cell.com] it seems to be Alphafold and the structure it reveals seems vital to being able understand how PHGDH could be interacting biologically and therefore be a critical part of the discovery, so a shame the uscd.edu article doesn't mention it by name.
Re: (Score:2)
Note that scientific papers have always have had to do some marketing. You don't get those grants to do the science without generally engaging in some fluff marketing along the way.
Re:Funny and sad (Score:5, Informative)
Re: (Score:2)
Re: (Score:2)
Even LLMs are better than many people are willing to admit. But that's what you expect when people feel threatened.
Irritating (Score:4, Informative)
Turns out it was AlphaFold to determine the structure of the folded protein, and ChatGPT was used to grammar check the paper. https://www.cell.com/cell/full... [cell.com]
Re: Irritating (Score:4, Interesting)
Did you know that AlphaFold uses the Attention mechanism, which is the breakthrough that allowed ChatGPT to solve context-sensitive grammar, thereby supporting the idea that natural language processing is AI-complete: once you solve NLP, you can solve every other AI problem?
Re: (Score:2)
OK. But a Turing machine is Turing complete. It's just a horrible way to program. Even assembler (think MIXX or I6502 assembler) is easier.
Similarly, for many problems that can technically be solved by AI, some other approach is better. It's NORMAL for the more generalized approach to be clumsier and more expensive.
Saying "you can use this approach to solve every AI problem" is not equivalent to saying "this is a good approach to use in solving every AI problem".
Re: (Score:2)
You're right that they could have explicitly said that, but really there's no other game in town.
Don't let genetics be your full stop. (Score:1)
Too often genetics and a hand wave away.
But genetics doesn't operate in a vacuum, always interacting with the environment.
It's just a matter of scale and perspective.
Re: (Score:2)
AI will remove medical care for millions of people (Score:1)
https://developers.slashdot.or... [slashdot.org]