Could AI and Automation Find Better Treatments for Cancer - and Maybe Aging? (cnn.com) 22
CNN looks at "one field that's really benefitting" from the use of AI: "the discovery of new medicines".
The founder/CEO of London-based LabGenius says their automated robotic system can assemble "thousands of different DNA constructs, each of which encodes a completely unique therapeutic molecule that we'll then test in the lab. This is something that historically would've had to have been done by hand." In short, CNN says, their system lets them "design and conduct experiments, and learn from them in a circular process that creates molecular antibodies at a rate far faster than a human researcher."
While many cancer treatments have debilitating side effects, CNN notes that LabGenius "reengineers therapeutic molecules so they can selectively target just the diseased cells." But more importantly, their founder says they've now discovered "completely novel molecules with over 400x improvement in [cell] killing selectivity."
A senior lecturer at Imperial College London tells CNN that LabGenius seems to have created an efficient process with seamless connections, identifying a series of antibodies that look like they can target cancer cells very selectively "that's as good as any results I've ever seen for this." (Although the final proof will be what happens when they test them on patients..) "And that's the next step for Labgenius," says CNN. "They aim to have their first therapeutics entering clinics in 2027."
Finally, CNN asks, if it succeeds is their potential beyond cancer treatment? "If you take one step further," says the company's CEO/founder, "you could think about knocking out senescent cells or aging cells as a way to treat the underlying cause of aging."
The founder/CEO of London-based LabGenius says their automated robotic system can assemble "thousands of different DNA constructs, each of which encodes a completely unique therapeutic molecule that we'll then test in the lab. This is something that historically would've had to have been done by hand." In short, CNN says, their system lets them "design and conduct experiments, and learn from them in a circular process that creates molecular antibodies at a rate far faster than a human researcher."
While many cancer treatments have debilitating side effects, CNN notes that LabGenius "reengineers therapeutic molecules so they can selectively target just the diseased cells." But more importantly, their founder says they've now discovered "completely novel molecules with over 400x improvement in [cell] killing selectivity."
A senior lecturer at Imperial College London tells CNN that LabGenius seems to have created an efficient process with seamless connections, identifying a series of antibodies that look like they can target cancer cells very selectively "that's as good as any results I've ever seen for this." (Although the final proof will be what happens when they test them on patients..) "And that's the next step for Labgenius," says CNN. "They aim to have their first therapeutics entering clinics in 2027."
Finally, CNN asks, if it succeeds is their potential beyond cancer treatment? "If you take one step further," says the company's CEO/founder, "you could think about knocking out senescent cells or aging cells as a way to treat the underlying cause of aging."
So none of us are going to have jobs (Score:1)
I mean there are probably solutions to that problem but God knows we're not going to come up with them. And if we do some stupid little bauble or petty mortal panic will distract us until we're dead.
Have fun in the comments! List your favorite moral panics!
Re:So none of us are going to have jobs (Score:5, Informative)
Shut up and let it find the cure. Your moral panic is blocking it. I'm not sure what you want, humans have tried to cure cancer for decades. We're not going to be able to crack it without AI and robotics, there's too much complexity. I mean, a cure is possible. In fact I'm in this field, there are many paths to being able to cure it .. and in fact the cure cost would likely be extremely cheap once the massive R&D is paid off. For example, here's one:
Cancer is a disease caused by DNA changes that cause the cell to divide uncontrollably, while simultaneously evading the immune system. So the obvious way to cure it is to identify and destroy only the cancer cells. Practically this is very difficult to do. The surface of a cancer cell, is very similar to the surface of a normal cell. There may be a few differences not the surface, but most of the differences are inside the cell. So we need a therapy that can detect what's inside the cell and based on what it finds, destroy the cell. The difficulty we've encountered is that you can't just transport that "detect and destroy" system into the inside of a cell.
The more advanced your "detect and destroy" system is, it is most likely a large macromolecule of some kind. That has to be transported into the cell by some sort of a carrier. The problem is we don't have good carriers, they either don't get into cells efficiently (LNPs for example) or the carriers are terrible at evading the human immune system (Adenoviral vectors, for example).
So anyway here's the history and future of cancer treatment.
1. Surgical removal -- fails when the tumor has spread or is in a location where surgery can't reach .. fails when the cancer mutates and modifies or hides the marker.
2. Chemotherapy -- uses small molecule or radiation that kills all dividing cells. Has side effects, and fails when the cancer cells become resistant and more tolerant than normal cells.
3. Monoclonal antibodies == target a specific surface marker on certain types of cancer cells
4. Targeted small molecules -- small molecules that target cancer-specific internal proteins that are essential to the cancer cell's survival. Fails when the cancer cell evolves its efflux pump mechanisms (such as P-gp or MRP1) or utilizes/creates a different pathway.
5. Immunotherapy -- boosts the immune system by helping it better lock onto surface markers on the cancer cells. Fails when the markers change or cancer uses a different immune evasion method.
Future -- deliver a payload into the cell and have the payload detect the cancer state by checking for the existence of DNA mutations, cancer=specific RNA, or proteins. If such a state exists, then destroy the cell .. otherwise do nothing.
The problem with the "future" solution is we still can't design a delivery system that is 1. non-toxic to normal cells 2. can efficiently enter any kind of cell. I know for a fact it's solvable as we have molecules that kind do one well, but come up short on the other. For example, we've been using Lipofectamine in the lab for decades which is amazing at "entering any kind of cell" and delivering RNA payloads .. but it has an unacceptable toxicity rate. When I say unacceptable, it's still pretty damn low like a single digit percentage of cells die .. that's fine in a cell culture plate .. but you don't want that in your body.
My point .. if AI can design a large payload (500 kDa, that's big but fuck it we're fantasizing) delivery system that is non-viral (therefore immune evasive) that can enter every type of cell we'll be 99% of the way to a cure (the other parts are trivial). Furthermore, the delivery system may need to have 3 or 4 "types" such that the cancer cell's efflux mechanism can't easily evolve to identify and expel it. It isn't easy in fact we've wasted a lot of resources at my lab and only have a few possible candidates that might work ,, that too with a few years of fiddling.
Re: (Score:2)
Re: (Score:2)
What's your suggestion? Ban AI and keep humanity at its current level? If it were 1960, you'd be asking for a ban on calculators. if it were the year 1600, you'd ask for a ban on textbooks or reference books since they take consultant jobs away from experts in various fields.
Re: (Score:2)
Yes (Score:2)
AI can do a lot with the data that we already have. That's all it's good for really. Give it a bunch of data and it can make correlations. Probably the biggest one it will find is that scientists like to collect data on people that are likely to have cancer. oops
It's tar! (Score:2)
Re: It's tar! (Score:2)
The rare superman 3 callback.
Re: (Score:2)
If they add tar as the additional component, then I'll be able to implement my special software to take a fraction of a penny from every transaction!
I'm sure the AI is already able to run tar. Also, probably gzip by this point.
Terrible Headline. (Score:4, Informative)
Could AI and Automation Find Better Treatments for Cancer - and Maybe Aging?
Not "could", "are", and the answer is a resounding "yes". Machine learning has been spitting out novel and potentially useful drug molecules for years now and automation has sped up DNA sequencing immeasurably, not to mention making it cheap enough that you can order a machine online and get something that will fit on your workbench.
More recently, AI has been applied to radiology - mammograms, specifically - and shows promise in detecting cancers smaller than a human doctor working alone could. Though this obviously doesn't improve the actual treatments used earlier diagnosis invariably leads to a better prognosis.
Knowing next to nothing about medical research or AI I wondered about applying LLMs to proteins; after all, DNA is just a language with four letters. But then I remembered Folding@Home and when I read a little about it I found that it is (or at least was) very difficult to predict how a protein folds itself up when just looking at the DNA sequence, and that it's the shape of a protein that determines what it does. I've no doubt that there are people around the world applying AI to this problem right now.
Speaking of F@H, is my memory playing tricks on me or was there an older version/similar programme where you could try to fold a 3D model of a protein yourself? This would have been fifteen years ago, easily.
Re: (Score:3)
Knowing next to nothing about medical research or AI I wondered about applying LLMs to proteins; after all, DNA is just a language with four letters. But then I remembered Folding@Home and when I read a little about it I found that it is (or at least was) very difficult to predict how a protein folds itself up when just looking at the DNA sequence, and that it's the shape of a protein that determines what it does. I've no doubt that there are people around the world applying AI to this problem right now
Alphafold https://en.wikipedia.org/wiki/... [wikipedia.org] https://slashdot.org/index2.pl... [slashdot.org]
Re: (Score:2)
Yes, a terrible headline but not for the reasons you list. Could AI do that? Yes, anything's possible. Could humans also do that? Yes.
Will AI do that? What difference does it make? We won't pay for medical care for serfs, and if you're not a billionaire you're a serf. Cancer treatments will be a billionaire's entitlement, who cares what they get.
Re: (Score:2)
Hold on there! You're making an important assumption there but not mentioning it: namely, you're assuming that there's at least one answer out there for AI to find. Yes, AI can find one if there's an answer to find, but until it does find it we won't even know if there actually is an answer, and that's why the headline is phrased like that. Unlike you, whoever wrote that headline isn't making unjustified assumptions. Frankly, I hope there is an answer and t
Re: (Score:2)
The difficulty here is the amount spent on AI and the like vs. other research- which will have the most efficacy?
The promises of AI are far and wide, as if to justify spending even more at the expense of other things.
Yes, new tools are helpful in many disciplines, but that's not what is being sold here.
Re:Terrible Headline. (Score:4, Insightful)
> The difficulty here is the amount spent on AI and the like vs. other research- which will have the most efficacy?
Lets see. Protein folding using "other research" methods for a single protein takes about half a year and costs about 200 000 dollars.
Time it took to develop AlphaFold2 was about 3 years. I am not exactly sure about the cost, but considering Deepmind budget being something between 500 - 1000 million per year, it should be a lot less than 3000 million dollars, so lets go with that. Once the AI was ready, it solved 200 000 000 proteins.
So lets calculate the cost for a single protein:
Traditional research: Half a year, 200 000 dollars
AI: About 0,5 seconds, 15 dollars.
And it is not that AI just solved a lot of proteins. AI created also a solution for the protein folding. It is several decades old problem that has been holding back biology research and medical research. Pretty much instantly after this, our understanding of one part of human cell rice from about 20% to 80%, simply because with the folding we could pretty literally see how the cell structure works.
And this was just the start. Wait 2 years and then compare the cost on drug research to AlphaFold3. Because it isn't just that you can invent new drugs that cure incurable deceases, you can also test those drugs against EVERY human protein to see if they have side effects or not. Which means that you know how the drug works, even before you start testing it with animals.
And once the AI is ready, it won't just make one drug, it will make all the drugs, for all the deceases. Can you beat that with "other research"?
Re: (Score:2)
Hmm...
Folding@home [sciencedirect.com] has already a list of accomplishments for nominal costs.
Further, picking specifically a Nobel prize winning AI as demonstrative of cost, when Forbes places the total spending for AI next year alone at a quarter trillion is suspect to say the least.
That's a lot of research dollars, but hey you accomplished what was already being done for free and with no restriction on copyright. I'm positive Alphabet spent that money strictly out of benevolence and will do the same.
But as we aren't discus
robotic experiment (Score:2)
From what I can see, they have some robotic machinery that takes arrays of antibody samples and inoculates a lot of petri dishes. Then they probably have a camera system that reviews the results in the dishes and decides what tweaks to make to their designer antibodies. I can definitely see how that would improve turnaround, and lots of iterations could lead to some things to try on mice.
Re: (Score:2)
Correct, I think robotics and lab automation could have a bigger impact than AI.
Re: (Score:2)
That is not true. Consider that your problem is winning in the game go and you have two systems:
A) Robot that moves the peaces on thousands of boards, testing thousands of alternatives and picks the best alternative from those.
B) You have an AI that very quickly comes up with almost perfect solution.
Even if you can try thousands of alternatives and pick perfectly the best one from those, you are still leaving out astronomically large number of moves that you won't be testing. This is why IBMs Chess engine c
Call me when they do (Score:2)
Otherwise you're just trying to griff people into investing in you.
No (Score:1)