Ask Almost A Doctor: Papal Blessing Edition
Edition Two
If you have questions, you can email me at eryneym@gmail.com, DM me on Twitter or Substack. Or put them in the comments below!
Also, none of the below constitutes medical advice. (Seriously. This is not medical advice - Ed.)
Enjoy.
Algobaker @algobaker
How will the economics of capsid designs end up working out? Will there be libraries of patented designs, and any new group designing a new payload will have to choose between paying the tax to use an existing capsid they see can already do the job, vs paying to do the experiments necessary to patent-bust it?
Great question. Obligatory mention that I’m a former employee and current shareholder of Dyno Therapeutics, which (I personally think) is setting the frontier of capsid design.
Now that I’m done talking my book, let’s turn to your question. Engineering genetic delivery vectors today is mostly about picking your priorities. Generally what people try to select for is a virus that can go to specific organs / regions of the body, but there’s also considerations like avoiding the immune system, avoidance of the liver (we call this detargeting) or to a lesser extent, production efficiency. An entire field exists around trying to manipulate these viruses using AI models and fancy protein engineering, but it’s not really something that can be done without intense experimentation with very long feedback loops that take a lot of time.
How can we shave time? Well, AGI of course! This is because it’s probably the only real avenue towards avoiding the requirement of testing hypotheses in the lab. If you can avoid the experiments, you save both time and money, and thus open up a ton of options for yourself.
Until we have superhuman intelligence my sense is your need to pay the tax to the current incumbents for existing capsid IP comes down to how much you’re feeling the AGI, and secondarily, how organ-specific you need a capsid to be. Though the ideal world is one where you have a 1000-fold better brain delivery capsid than, say, AAV9 (the best “free” natural variant for the brain), you need to ask yourself whether you can get away with 5x? What about 2x? In that case it’s possible you can design your own capsid with a combination of off the shelf ML models, a good cloning and production pipeline and some non-human primates. I don’t know what it would cost you (at least $250k if I had to guess) but it would definitely take you time.
It’s worth mentioning here that when I say AGI I mean legitimately AGI, not models that are 10% better at writing code. I personally think the only real way it becomes cost efficient to engineer your own capsids instead of paying for it is that the field gets access to protein ML models that can tell you with very high confidence how a protein will behave in a zero-shot manner. Despite what you read online, we are not there yet, and we won’t be for many years.
The downstream consequences of this timeline is an exercise I will leave to the reader.
David Dales @d2dev_
Now AI is out and public figures are telling me more hospitals are hiring more doctors to use the AI - can you confirm or deny with data? I heard x-rays and MRI scans largely use AI to detect issues these days. I think it was in the last 5 minutes of the recent Lex Friedman/Jensen Huang podcast.
I’m going to take your question to refer to radiology, as that is what people generally mean by AI replacing doctors. Radiologists are actually in such high demand that they could easily out-earn neurosurgeons if they decided to work more than a few months per year. But how can this be true? Well, let’s start by first addressing the misconception that AI is replacing radiologists.
NVIDIA CEO Jensen Huang, and also Anthropic CEO Dario Amodei, are both incredibly smart guys who have gotten this one wrong completely. I won’t comment on how that happens — keep in mind they both have reason to push the narrative that AI can do complex jobs easily today — but I will say that this is a pervasive myth in the tech community. I will put it very plainly here: no hospital is replacing radiologists with AI today. While the range of what AI models can do is growing (see here this study on a new neuroradiology model from Michigan), the skillset is still incomplete, and thus hasn’t changed the job landscape at all.
The hiring effects you see with radiologists are actually the result of something entirely unrelated to AI, and that is the rise of advanced imaging in medicine. Previously (and by previously, I mean like 50 years ago), doctors put a lot of weight in their clinical intuition and the art of the physical exam. Sadly, that is a fading skillset, but we can directly tie it to the rise of on-demand CT and MRI in healthcare. Have an ear ache? Head CT. Strange lump? Ultrasound. Think you tweaked your knee? Let’s get you an MRI. All of these things were at some point diagnosed from physical exam findings, but now get imaging. The result is that we get imaging on way more patients than we used to, and that amount is further increasing as new modalities get added.
I can’t read the future, so I won’t make a prediction here about eventual AI capabilities in radiology. The point is though that I think the hiring happening in healthcare is very much real today.
Claire Goldsmith @c_goldsmith
What is going to happen with monoclonal antibodies over the next two decades? Everyone talks about costs coming down and this not being a particularly attractive part of the market long-term, but it’s doing very well for big players today (J&J etc). Other than patent cliffs, what do you think actually manifests that cost curve compression? Seems much more like a manufacturing problem than a design problem. Also, which comes first, broad use of mabs for more disease areas outside indication or major decrease in cost of manufacturing?
Are you feeling the AGI, Claire? Monoclonal design is getting better thanks to advances in AI-enabled protein design — faster than most proteins right now, I’d say — but honestly, target selection seems like more of the rate limiter here. Everyone seems to be tackling the same exact ideas. Structure models like AlphaFold or RosettaFold or Chai seem poised to help the design of drugs that act on established targets with improved efficacy / potency (referred to as me-betters) but they don’t really help you pick out novel idea space. In theory, that’s where AGI helps. I am not convinced we have it yet, though.
There’s another element to the economics here which is that the arrival of biosimilars (“generics” for antibodies) drives the price of monoclonal antibodies down significantly, upwards of 80%. When this happens doctors become more willing to prescribe a particular biologic for off-label use. I expect that to happen more and more, especially now that the FDA has said their goal is to make biosimilars easier to get through the pipeline.
If your question on manufacturing is one of cost, I would argue that’s not really the problem. Right now most monoclonals can be made around $100/g. If we look at the cancer buster Keytruda, which is dosed at ~2mg/kg every few weeks, you’re looking at a max of around $500 in terms of production costs. The gap between that number and the $150,000 price tag is owed to amortization of R&D and clinical trials.
Claire Goldsmith @c_goldsmith
Growing organs…. Is this working? Will we be able to do it? What problems does it actually solve? Transplant success rates after the 1-year mark are not improving, and I don’t think organ supply is the problem.
If you’re referring to growing whole organs, we’re quite far off, so I think it would be unwise to be all-in on this. There are other options though, like xenotransplantation – taking organs from other organisms (namely pigs) – which are kind of getting there. NYU has a trial for kidney transplants from pigs and as of today, the longest survival time is 9 months. Again, not bad, but as you point out, not enough. This is a little above my pay grade, but my understanding is that the main way we humanize pig organs is to eliminate endogenous retroviruses within the animal that immediately activate our immune systems if they get transplanted. Unfortunately, there seems to be some antigen that has yet to reveal itself, which ends up the same way as many human-to-human transplants — rejection.
Whole organs are a difficult business, but patches seem viable. Lots of companies are working on this, new and old. I wrote about one last year, Polyphron, but there are plenty of others. The goal of these approaches is to swap out broken bits of an organ. Also not there yet.
You didn’t ask, but I think it’s kind of cool that since our last edition, the Pope issued an official decree that Catholics are able to accept pig organs according to the Written Word. America’s Pope supports American biotech. Annuit cœptis.
waitingonyou @Imyouropnow
Why are there more investments in AI innovation at the bench rather than the beside? Do you think it might be possible to infer immunotherapeutic effects (w/o drug perturbation experiments) through, for example, cytokine-symptom effects at the bedside?
Your question reminds me of a fantastic book I read a few months before the COVID-19 pandemic, The Great Influenza. It’s about the Spanish Flu (or Kansas Flu, IYKYK). The flu accelerated medical science, but split the process of discovery between the bench and the bedside. There became a specialist class of researchers whose whole thing became studying biological phenomena independent of the treatment of patients. This is great, but it did have cultural consequences. Doctors don’t really do the physician-scientist thing like they used to. They rely on the biological sciences to be the engine of discovery and while some patient-facing physicians try things in the clinic, our healthcare system works well by having doctors mostly implement things once they’ve been established as safe and effective in small numbers via trials.
So, why is there no AI innovation at the bedside? Well, it’s not immediately useful. AI-powered tools like OpenEvidence are great for distilling dense medical literature to a specific question, but that’s not really the same thing as innovating, is it?
Now, I do think there is a lot of useful biological data to be gathered from the clinic for those with the stomach to figure out how to get it. A large part of the limitations are the fact that we have very poor measurement tools, which is why I spent some time working on this at Caltech. But the tools largely still need to be built. There probably are insights that can still be gathered, though, so if you’re an engineer or scientist who wants clinical data, try connecting with a clinician.
Towards your specific comment about cytokine-symptom effects, I expect it’s mostly just that doctors are waiting for the science to clarify what they should do. AI is best served doing that, because as things go today, that’s not really the job of doctors. An interesting question is whether AI will enable that to happen, though. That’s one whose answer has yet to reveal itself to me.
Ashlee Vance @ashleevance
If I’ve already had shingles, should I take the vaccine anyway, too?
The simple answer is yes. The longer answer is definitely yes.
Shingles is the result of varicella zoster — the virus that causes chickenpox — staying dormant in your nerves after you clear the initial infection. Because our immune systems wane in efficacy over our life, the virus generally gets reactivated resulting in what I’ve heard described as the worst pain imaginable. If you’re unlucky, you can actually get it again at some later point — having shingles once doesn’t mean you’re set for life. It’s recommended that anyone over 50 gets it. As a Sensitive Young Man of 29 Years Age, I naturally haven’t gotten the shingles shot, but I’ve heard it’s quite painful. Still, less painful than shingles!
I’ll add that a study in December 2025 in Wales shows that, while the shingles vaccine doesn’t stop dementia once it gets going, it can slow its progression down and even prevent new cases in the vaccinated. They use two different cohorts to demonstrate that the effect is real and is not the result of weird selection effects. Interestingly, they show that this effect is stronger in women than in men.
So, Miss Ashlee, I’d get your vaccine.
Happy Friday.
 | A guest post by
|