Ask Almost A Doctor: A Functional Health Joint

Edition Four

If you have questions, you can email me at eryney@corememory.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.)

And as always, thanks to Kylie Robison for editing. Please check out her directorial debut with Core Memory when it comes out.

Thanks for reading.

Jacob Keeton (@jacobkeeton via substack)

Trying to think of how I could make this question about space.... Maybe something with astrobiology?

How about hyperspectral biology? This is something I learned about from my friend Niko McCarty, a wonderful writer and brilliant biologist. The term captures a broad set of things, but the general idea is to create methods of viewing biological phenomena using satellites. For instance, could you create a system capable of detecting explosives hidden in the soil?

An MIT study showed that it’s possible to engineer bacteria that, when sprayed into soil laden with TNT, would make light that can be visible with satellites in space. You can maybe imagine the use for something like this given the current state of the world, but there are benign use cases too, particularly in agriculture. Such a system of bacterial biosensors can track heavy metal or toxins in water supplies, monitoring fertilizer runoff or even helping farmers know if their soil needs nutrient repletion. I’m not a horticulturist so pardon my lack of imagination, but I’m sure intelligent people trained in those areas can think of some uses.

Also, you should read Niko’s work.

Siddharth Sethia (@siddharthsethia via substack)

Do you think binder generation is still an interesting problem to work on? If so, what types of targets or types of binders are underserved? If not, why do you say so?

I’ll also point at another friend here, Abhi (aka Owl Posting), who convinced me that actual design of protein binders – a field specific term for a protein that has high affinity for something else and that’s how it exerts its effect – is more or less a solved problem. BindCraft out of MIT has sufficiently demonstrated that making novel, functioning binders is no longer a bottleneck. In that sense, yeah it’s probably not an “interesting” problem to work on. However, what do you make binders for?

ML in bio has advanced to the point where generating and selecting proteins that have desired properties (high affinity, increased stability, etc) is quite good with off the shelf tools. When I first started my biology career in 2014, this was not true, but things started to improve with Google’s AlphaFold in 2020. The reason this hasn’t resulted in a tidal wave of new therapies is that design of biological molecules is only one part of the problem. You have a bunch of nails – awesome! What are you going to do with them?

The next wave of biological research orgs are going to focus on improving the loop that is both target discovery and binder design together. So far, I’d say we’re a ways off from solving the discovery part.

Gviv (@gviv on substack)

What are the main physiological mechanisms that the body uses to prioritize and reallocate nutrients when they are in short supply?

So many places I can go with this one.

Let’s maybe just focus on the basics of energy balance. At the highest level, your body can use fat, protein or carbs as energy sources. Carbs get broken down into basic molecules of glucose, which your brain uses a lot of, and your liver can store in the form of glycogen when you have excess. Fat and protein can go a similar way – storage, or use. What happens when you’re fasting?

Your body will break the glycogen down into glucose to meet the most immediate energy needs, but that gets depleted pretty quickly. It’ll then start making new glucose by breaking down either fat or protein. Generally, the body starts with fat, but if you don’t have much of it, there’s no choice but to go for protein. On a long enough time frame protein ends up going, and at that point you’re in trouble. This is all stuff that you can learn in a bio 101 class, but it’s the fundamentals that make up a ton of nutritional and exercise science.

An additional, more biochemically focused framing can be spotted by paying attention to the actual cycles involved in making these things. Each of the processes I described above involve dozens of intermediate steps that involve an input substrate, an enzyme, and oftentimes a co-factor like vitamin B12 or NAD+ and so on. Cells in your body manage the broad pathways by monitoring the build up or absence of individual parts of the cycle. Too much ATP, and the body will slow down glucose use. Too little, and it’ll start making more glucose.

These feedback cycles have so many distant regulation points that it makes much of the discourse around what vitamin or mineral or supplement to take just impossible. Everything is plausible under the sun, but much of it fails to hold up under experimental scrutiny. Here is a map if you’re curious.

Source Evans Love https://en.wikipedia.org/wiki/Metabolism

This is just metabolism, by the way.

The Middle Aged (via substack)

I’d be curious to know what the best ways are for young educated patients (who are not medically trained) to seek out the latest research advances related to their conditions.

I grew up with a rare blood condition, idiopathic thrombocytopenic purpura (ITP). At the time of my diagnosis – age 13 – I knew basically nothing about biology or health. However, the internet is great, and I got to reading about some of the basics of the blood and immune system, which kicked off what would become the last 16 years of my life. Something I used to do was occasionally search up and read trials for my condition. I didn’t really understand much of what was in there, but that’s OK. The first time I actively read a paper I only understood 1%. The next paper, maybe 5%. After the 10th paper in the field, I was following most of the story. Again, I was a dumb teenager, and you could do much better than those numbers starting today.

What I’ve observed in my (limited) clinical experience is that patients who engage with their condition can participate in as much of their care as they’d like. With LLMs, this has become much easier. As an example, once you understand the basics, you can keep up to date on trials by setting PubMed email alerts for keywords. Depending on how invested you are, patient advocates are welcome at many conferences, too, but those tend to be a little pricier.

Sri Nandan Gondi (@srinandgondi2 via substack)

I’ve seen the Bio space, especially ML Bio, heavily focus on better drugs, therapeutics, etc. But how much progress can we make in preventive healthcare (food, lifestyle, environmental chemical exposures, pollution, aging, etc) and how can ML help in these areas?

Medicine and lifestyle tend to be viewed in opposition, but they really aren’t. For instance, we know obesity can by itself cause many different types of cancer, such as with endometrial cancer. Though we now have medication to manage weight in the form of Danish Peptides, clearly diet and exercise are sufficient for many people.

There’s now plenty of evidence that environmental exposures can cause cancer, as was the very unfortunate case of Teflon (and PFAS in general). You don’t need me to tell you that cigarettes aren’t good for you, or that drinking in excess can cause a litany of health problems. The reason we know all of these things is because large retrospective studies analyzed the effects of many now-identified carcinogens on populations and were found to have bad outcomes. These served as the basis for prospective trials, and ultimately cementing the common advice: don’t smoke, drink, or eat red meat.

Where ML can be helpful is making use of all the monitoring data that is cropping up. Millions of people are having their hearts monitored by their Apple Watch or Whoop. Thousands are getting more routine blood testing via various services like Function Health. The UK Bio Bank, which has followed 400,000 citizens for a couple decades and captures a ton of different data modalities, is a great example of what’s possible. Imagine what we learn with an additional order of magnitude worth of data.

Trina Patton (@trinapatton698090 via substack)

I’ve always wondered how doctors figure out what’s going on when multiple illnesses can have the same symptoms. Is it mostly pattern recognition, testing, probability, or something else that guides treatment decisions?

Pattern recognition is a big part of it. Even mediocre doctors are very good at catching the common stuff. As you’re suggesting, the challenge is what to do when there are multiple things going on, or whether someone is a medical zebra.

I will say that medical students actually do learn a tremendous amount about identifying ultra-rare diseases. For the US Medical Licensing Exam we’re expected to know how to differentiate various one-in-a-million immune deficiency disorders (Whiskott-Aldrich Syndrome, Hyper-IgM, IgA deficiency, SCID, CVID and many, many more), down to the exact dysfunctional genes that cause them. As you enter practice, though, you see less of them, and that knowledge atrophies. That’s why generalists rely on experts when they don’t know what’s going on.

For example, if you go to your primary care physician as a 30-something man with recent multiple bouts of diarrhea after trying this year’s mystery meat in the McRib (it’s not back, by the way), they can probably help figure out if you just have a GI bug. If it persists for a few weeks, it’s unlikely to be that, and they are likely to order some blood tests that might help whittle things down. HIV or hyperthyroidism can cause diarrhea, but so can celiac disease. If all of that is negative, you’re off to the GI specialist for a colonoscopy and whatever else they have in mind.

The American healthcare system is actually exceptional at solving problems, despite what you read online. But you do pay for it.

Jennifer Yates Zilliac (@thisisthemountain via substack)

I’ve been trying to solve some issues with sleep and energy for well over a decade, and I keep finding there is one more piece to the puzzle. I’ve been frustrated that my primary care docs have not partnered with me to solve problems. I’ve now found that partner in AI, and the approach is functional medicine, which is not what my primary docs are doing with me. Please tell me what you see in terms of a shift toward functional medicine and AI.

I’m sorry you’ve been dealing with these issues. They seem disruptive, and I hope you’ve arrived at something that is closer to normalcy.

Functional medicine seems like something that feels uniquely resolvable by AI that traditional healthcare isn’t. Continuous monitors capture tons of data; digital food logs keep track of what’s going in your body; and blood testing is becoming more available outside of the usual medical infrastructure. The catch is that the onus is on you, the patient, to provide context.

If you are wearing your Apple Watch, and you experience a period of accelerated heart rate at 11pm, your device can’t really tell you what’s going on. Maybe you went for a late night jog, and it can somewhat intuit that from the accelerometer, but it doesn’t know that what actually happened is you drank a couple cups of green tea with dinner at 9pm. In that instance, AI isn’t providing anything for you. Then again, neither is a doctor. But an invested human is more likely to ask you about your sleep habits generally, which might lead to surfacing of this link.

I used a Whoop for some time to keep track of things, but honestly I found it overwhelming to have to fill out my daily journal of whether I had 1 cup or 1.5 cups of coffee and at what time, whether I flew on a plane, whether I ate dinner within 2 hours of bed, and so on. Figuring out a way to keep track of the granularity in our day seems like the next thing for these companies to do. Maybe something like Meta glasses will help with that. Though I hate the idea of Apple acting as a Big Brother, you can also see a world where Apple can make use of the fact that it has access to your texts, calendar, flight details etc to automatically link to your Watch data. They just gotta get their act together with AI.

In any case, the best solution involving functional health is one where your health data can be understood in the context of you. Whether that’s with AI-integrated devices or by talking to a functional health specialist is up to the individual.