AI could help lead to an Alzheimer’s cure

Lana Del Rey’s “Young and Beautiful” starts with piano keys, church bells, violins soaring through the first verse. Then a drum roll, and Lana belts out:

Will you still love me when I’m no longer young and beautiful?
Will you still love me when I got nothing but my aching soul?
I know you will. I know you will.
I know that you will.

You can read those lines and roll your eyes at the cliché. Cringe at her saying it out loud in front of a microphone. She wrote it for a movie, literally for The Great Gatsby! But when that song last played in my headphones, walking in the sun to the park, I heard Lana’s voice, her yearning, her singing one of each of our (my?) deepest fears. Reassuring herself that it will be fine when in fact she cannot know the future… My soul ached too. There is a reason “Young and Beautiful” has 1.7 billion streams on Spotify. Lana aimed big at a universal human experience. She did not miss!

If you don’t know that song, or it’s not your style, think of what you feel when Chris Martin from Coldplay sings: “I will try to fix you”. When Robyn sings “I’m in the corner watching you kiss her. I’m right over here, why can’t you see me? … I keep dancing on my own.” Adele: “We could have had it all…” Kendrick: “Do you hear me, do you feel me? We gon’ be alright.” Bruce Springsteen: “Born in the U.S.A.​ I was born in the U.S.A.​”

Each of them is aiming big, they are aiming simple, they are aiming universal – or, in Bruce’s case, universal for 340 million people. I can’t bring myself to cringe at any of them, I don’t care that they’re cliché, because they hit.

So as much as I roll my eyes at tech people (of whom I was one, 2015-18; all hate is self-hate, after all), at the self-importance, the over-simplifications, at missing some of the point of living… artificial intelligence is real, it’s simple, and it hit. You can, right now, talk to a machine that will teach you things you want to know. This one, this one, or this one. You can ask them about Bruce Springsteen’s life, or about black holes, or about what AI progress might mean for medical progress, and you can ask follow-up questions until you understand to your satisfaction – and nowadays you can ask for links and references to check if they’re making something up (which they do).

Other AI systems can predict how most proteins fold up in your cells, plus hallucinate new proteins that bind and neutralise bad stuff – e.g. toxins in snake venom. That used to be a dream, now it is real.

“AI” is cliché, so cliché sometimes it feels meaningless and draining to talk about, but… I cannot cringe at those achievements. A machine can teach me things I want to learn, and another machine can hallucinate drugs that work in the lab. Those are simple, and big, and stand for themselves. Those did not miss!

This post argues AI progress is a driver of medical progress. If you already believe that, and wonder instead whether AI will change all of science suddenly, read this version instead.

AI is already driving some medical progress

OK, if AI’s making new drugs, how come I’ve never taken a drug AI made? How come I still get sick sometimes, how come my coworkers have had back pains that mean they can’t sit in a chair for more than a couple hours, and how come some of my friends have lost months of their lives to depressions that they can’t make sense of while trying drugs that didn’t actually help them?

New drugs are coming, one or two are approved, but mostly they’re not here yet. The snake venom one isn't on the market, and it may end up not being practical. The design of the initial molecules to test is pretty magic-feeling, and is happening already – but the arrival of new drugs prescribed by your doctor will not be magical and sudden. There are a lot of things that can go wrong after a scientist, with the help of an AI, designs a good candidate drug.

But AI is driving several improvements in several areas of science, and each will contribute to discovery and development. Here is a list of things that have happened since 2020, which was the year deep learning started breaking through to biology with AlphaFold2:

Incremental improvements (these add up, making scientists more productive):

• Writing code for bioinformatics. Scientists generate a lot of data, and have to clean it and make sense of it. That takes time!
• Models have helped detect fraud in publications/datasets
• Models have helped improve microscopy methods, so we can witness more biological phenomena at higher resolution
• Other models are being used in gene sequencing, to reach higher accuracy for genetic variants in a sample

Discontinuous improvements (these allow scientists to do qualitatively new things):

• Predicting how proteins will fold from just their amino acid sequence
• The reverse: taking a folded protein structure and predicting what amino acid sequence led to it
• Designing an entirely new protein to bind an existing molecule tightly, just from the structure of the existing molecule. (A lot of medical solutions start by binding something tightly!)
• Diagnostics on a phone with no clinic visit/doctor needed. That means new people will get screened who otherwise wouldn’t have had access to a test, for treatable conditions they might not know about.

Then there are things that people in San Francisco talk about that are hard to adjudicate, because they haven’t happened yet. In particular, perhaps future powerful AI could be a co-investigator or scientist itself, or lead to a “country of geniuses in a data center” where a million copies of future AI scientists get spun up and start debating ideas with each other and running computational experiments. I do not believe that AI could end disease within 10 years on its own even if people race at that goal; head over to San Francisco if you want to read why.

Outside of natural medicines, every drug that now exists at some point was invented, and everything we now know was discovered

In 1985 we did not have treatments for HIV/AIDS; now we have many. In 2000 we did not have cures for hepatitis C; now we have many. In 2015 we weren’t sure what caused morning sickness; now it seems to be GDF15, a hormone produced by the foetus.

We do not yet have a cure for several cancers – but could. We do not yet have a TB vaccine that works in adults – but could. We do not yet have a cure for Alzheimer’s – but probably could.

That means the progress of science in society is of tremendous importance to all of us. Discoveries and knowledge do not appear by magic. They come from scientists doing good work with good equipment, sharing ideas at conferences, reading each other’s papers, and following their nose without having to make money from their research. AI tools are helping make scientists more productive and creative already, and more discoveries should follow.

That’s a little abstract. Let’s make it more concrete.

Can AI help cure Alzheimer’s?

Maybe! Before we know if AI could help, why haven’t scientists cured it already?

Why is there no cure for Alzheimer's yet?

There are bottlenecks galore: only so many scientists, only so much funding, difficulty in taking measurements of what’s going on in the brain, difficulties interpreting signals in the blood we thought came from the brain, and a surprising amount of scientific fraud in the field over the last few decades.

What would it feel like if there were a cure for Alzheimer’s?

You, your spouse, your friend, your child, or probably some combination, notice – your forgetfulness is getting worse. You go to the doctor. They ask you and your family a bunch of questions about your medical history, and what’s been happening at home recently. They give you some puzzles to solve on paper. They do some blood tests to rule out vitamin B12 disorders and thyroid issues. You slide into an MRI machine, and your hippocampus looks smaller than it should. You receive a "probable Alzheimer's" diagnosis.

The doctor prescribes you one dose of Protollin up the nose, to activate your immune system against amyloid proteins in your brain, and a six-month course of Memorvio and Cathespian. You pick the bottle up at the pharmacy, and put it in the freezer. Each morning at breakfast (habit makes it easier), you squirt some mist up your nose. The mist contains copies of two strings of RNA in lipid nanoparticles. Some of those particles make it through your olfactory nerve into your brain; there, some make it into your brain cells. The Memorvio gets translated by your cells’ machinery into Brain-Derived Neurotrophic Factor (BDNF) proteins, and the Cathespian makes cathepsin D enzymes that lead to degradation of your tau tangles little by little each day. The RNA strings are designed to shut off translation if either protein is over-expressed. The spray smells a bit like ginger.

Over the course of the six months, you gradually feel your memory sharpen, and spend more of the day with your mojo back. You go back in for an MRI, and sure enough, your hippocampus hasn’t shrunk any further. A PET scan confirms you’ve gotten rid of the equivalent of 5 years' worth of tau buildup. The doctor says: let’s do annual checkups, but otherwise you’re good to go.

That first paragraph happens to people every day, and the second paragraph was science fiction. The cure described does not exist, though similar-ish ideas are being researched. Writing that second paragraph made me uncomfortable, for the act of typing made it feel real for a few seconds, and I visualised the people in my family who are scared of aging and how they’d feel reading it. The reason it’s uncomfortable to type as if it were true is because it isn’t true, but it matters so much.

That type of cure is something that, if we do not go off the rails as a species, I do expect to be available in 100 years. We are lucky to be born now, not 100 years ago, and we work today so that our descendants can be luckier than us. The apartment I live in was built by people my age in 1910, and I did nothing to lay the brick foundation. Hopefully those of us alive now can invent an Alzheimer’s cure, and pass that gift on.

OK, so can AI help us get to a cure like that for Alzheimer’s?

I bet it can help, by trimming dead ends and boosting research into promising hypotheses.

• Drug design. I said “Memorvio gets translated by your cells’ machinery into Brain-Derived Neurotrophic Factor (BDNF) proteins”. If you were making a drug like that in the 2010s, one issue you’d face is that BDNF proteins 1) aren’t naturally that stable, and 2) don’t cross the blood-brain barrier easily. In the 2020s, you could use a mix of AI tools like AlphaFold, ProteinMPNN, and RFDiffusion to iteratively alter nature’s design, changing those properties while maintaining BDNF’s core functions, ending up with a sibling protein. Then you could encode that protein on a strand of mRNA (just like Pfizer/BioNTech and Moderna encoded the COVID virus's spike protein on mRNA), wrap it in a lipid nanoparticle, and optimise that nanoparticle so some of the package gets past the blood-brain barrier. There, a new drug candidate!
• Fraud detection. Currently society has to rely on Elisabeth Bik. That’s vastly better than not having Elisabeth Bik, but at some point she deserves a vacation. She has found thousands of manipulated images in publications that throw the papers’ results into question. Other groups are developing AI systems to scan the literature for doctored images and doctored data, either in existing publications or for use by journals when they receive future paper submissions. Had those tools been around sooner, 21st century Alzheimer’s research may have been built on stabler foundations.
• Detecting useful patterns in samples outside of the brain. More sensitive and specific blood tests for Alzheimer's could be helpful for patients in their own right. They may also, in turn, improve the designs of clinical trials for treatments, by allowing cleaner inclusion and exclusion of patient profiles and detecting improvements quicker.
• Modeling the brain. When I was growing up, scientists had mapped the brain of the worm C. elegans – 302 neurons, around 7,000 synapses. Science must have kept going because last year a group of collaborators released a map of a fruit fly’s brain, which is vastly more complex, with ~140,000 neurons and ~50,000,000 synapses. Perhaps scientists with better visual and mathematical intuition than me know how to make sense of that, but I’m looking forward to AI taking a look too. Flies and humans aren’t quite the same, but we share 60% of our DNA, including some genes involved in neurodegeneration. What causal models can the machines come up with?

What should we do if we want even more medical progress?

Depends who “we” means; I have a limited perspective based on my experience and life so far. (Hello readers from around the world! I hope some of these ideas are useful to you wherever you are. First up, check whether your country contributes to the Global Fund or Gavi.) I’m a dual US-UK citizen and live in the US, so tilting towards what I know better:

• Continue trying to apply deep learning to parts of medical research that could be improved incrementally or drastically by AI: in particular modeling cells virtually to high fidelity, modeling molecular dynamics, drug discovery (there are many chronically underfunded biotech startups with talented founding teams), and toxicology + off-target effects prediction before finding out the hard way
• Reform the NIH and NSF and increase total public science funding, through them and other mechanisms. That includes global research funding, e.g. NIH subawards and the Fogarty Center
• Reform clinical trials. Reform the FDA, and some other policy stuff in those links too that’s hard to summarise in a bullet
• Fix market failures in part by being more generous as a society
  • There would be no AlphaFold without $10 billion+ of public funding over 50 years, supporting PhD students painstakingly figuring out 100,000+ protein structures and adding them to the Protein Data Bank
  • Other machine learning models would get nowhere if it weren't for the billions of dollars spent on the Human Genome Project, that paved the way for it now to cost under $1000 to sequence a genome
  • There is no point in a leading American antiviral company developing a miracle drug that prevents HIV unless people who need it can access it
• That means rich people too, giving away more money to good things, burdening their kids less
• Public and private funders try to prioritise funding by health impact, since some areas of science are 10X+ less funded than others for no good reason. Swing for the fences.

I already support a bunch of those ideas, unrelated to AI. What should I take away from this piece about AI?

Basically: don’t cringe, some of it’s real, and it may help you if you’re a scientist, or help you in your job if you’re not. If you have a principled objection to using chatbots, I tend to sympathise with principled personal stances, so won't try to convince you out of it. But if you think AI is just not that good, then here are a few things you could try, in case they help you:

• If you are a scientist who writes code, try the o3 model in ChatGPT ($20/month), Claude Opus 4 ($20/month) or indeed Claude Code next time. The language models are a lot better than two years ago, and could save you some work.
• If you’re a PhD student who can’t code and you don’t want to take a year out to learn, try asking one of the chatbots to write some code for something you’re trying to achieve, e.g. “write a draft Jupyter notebook that plots this data on graphs”. Ask the bot to explain what each bit of the code is doing, so you can learn as you go.
• If there's a paper outside of your field you want to read but realistically will never get to it, upload the paper to Notebook LM to generate a hallucinated podcast with two AI co-hosts explaining its contents to each other for 15 minutes. Actually, start by uploading one of your own papers, or something else long that you've written, to see if they make any mistakes. It’s spooky.
• If you’re a scientist who wrote off structural biology when you looked into it pre-2022, you may have been right to do so then – and may be wrong to stick to that verdict now. Take another look, and play around with ProteinMPNN, AlphaFold, and diffusion models like RFDiffusion to apply them to your domain before the generalists come crashing through the gates…

For a better messenger than me, watch David Baker’s speech after winning the Nobel Prize last year, to hear some of the things they’re working on at the University of Washington, e.g. proteins to degrade plastic, proteins that do photosynthesis more efficiently, proteins to vaccinate you against multiple infections at once. The strangest part of listening to those descriptions is that I’m pretty sure some of them will work. And for most of them, you could not make much progress, at least not very quickly, before 2022.