How Matthew Might Is Using Computation to Fight Rare Diseases
Harry’s guest this week is Matthew Might, director of the Hugh Kaul Precision Medicine Institute at the University of Alabama at Birmingham. Might trained as a computer scientist, but a personal odyssey inspired him to make the switch into precision medicine. Now he uses computational tools such as knowledge graphs and natural language processing to find existing drug compounds that might help cure people with rare genetic disorders.
Might’s odyssey began with the birth of his first child, Bertrand, in 2007. Bertrand seemed healthy at first, but soon developed a cluster of symptoms including developmental delay, lack of motor control, inability to produce tears, and epilepsy-like seizures. For more than four years, doctors were unable to diagnose Bertrand’s condition. But eventually a technique called whole exome sequencing revealed that he had no functioning gene for NGLY1, an enzyme that normally removes sugars from misfolded proteins. Bertrand, it turned out, was the first person in the world to be diagnosed with NGLY1 deficiency—and as with so many other “N of 1” diseases, there was no known treatment.
After the diagnosis, Matthew and and his wife Cristina decided to used social media and the Internet to locate other patients with NGLY1 disorders around the world. Eventually the couple discovered 70 patients with the condition. Reasoning from first principles about the role of NGLY1, Might discovered that giving Bertrand a sugar called N-acetylglucosamine, a metabolite of NGLY1, helped restore his ability to form tears. (Around the same time Might, co-founded a startup that screened existing drugs to see whether they could treat ion-channel-driven epilepsy similar to what Bertrand experienced; the company was quickly sold to Q State Biosciences.)
Working with collaborators at the University of Utah, Might studied planarian worms that had been engineered to lack NGLY1, and found that those that also lacked a second gene had a higher survival rate. That meant one way to treat Bertrand might be to inhibit the analogous gene in humans, in this case a gene for an enzyme called ENGase. Might used computational screening to look for existing drugs that would be inverse in shape and charge to the catalytic domain on ENGase, and might therefore inhibit it.
He found more than a dozen drugs that were already FDA-approved. One was Prevacid, a proton-pump inhibitor sold as common over-the-counter medication for acid reflux. It turned out that as a previously unsuspected side effect, Prevacid is an ENGase inhibitor. Bertrand started taking the drug, and Might says it was one of the treatments that helped to extend and enrich his life.
Sadly, Bertrand died in 2020 at the age of 12. But by that point, his father’s work to apply computation to basic biology, and thereby speed up the treatment of rare disorders, had sparked a movement that will long outlive him. Years before, Bertrand’s story had caught the attention of the Obama administration, which invited Matthew to the White House to work on a range of precision-medicine projects. One was an NIH program called the All of Us initiative, which is collecting the genomes and medical records of a million Americans to search for correlations between mutations and health impacts. Might also launched a smaller pilot program called the Patient Empowered Precision Medicine Alliance (PEPMA) with the goal of repeating what he and Cristina had done for NGLY1 deficiency—that is, quickly diagnose the problem and identify possible treatments.
Might resigned from his White House role about one year into the Trump administration, then got an offer from University of Alabama to come to Birmingham to set up an institute to scale up the PEPMA idea. One project there called mediKanren involves using logic programming to highlight what Might calls the “unknown knowns” in the medical literature and identify existing, approved drugs that might treat rare disorders.
Please rate and review MoneyBall Medicine on Apple Podcasts! Here’s how to do that from an iPhone, iPad, or iPod touch:
1. Open the Podcasts app on your iPhone, iPad, or Mac.
2. Navigate to the page of the MoneyBall Medicine podcast. You can find it by searching for it or selecting it from your library. Just note that you’ll have to go to the series page which shows all the episodes, not just the page for a single episode.
3. Scroll down to find the subhead titled “Ratings & Reviews.”
4. Under one of the highlighted reviews, select “Write a Review.”
5. Next, select a star rating at the top — you have the option of choosing between one and five stars.
6.Using the text box at the top, write a title for your review. Then, in the lower text box, write your review. Your review can be up to 300 words long.
7. Once you’ve finished, select “Send” or “Save” in the top-right corner.
8. If you’ve never left a podcast review before, enter a nickname. Your nickname will be displayed next to any reviews you leave from here on out.
9. After selecting a nickname, tap OK. Your review may not be immediately visible.
Harry Glorikian: I’m Harry Glorikian, and this is MoneyBall Medicine, the interview podcast where we meet researchers, entrepreneurs, and physicians who are using the power of data to improve patient health and make healthcare delivery more efficient. You can think of each episode as a new chapter in the never-ending audio version of my 2017 book, “MoneyBall Medicine: Thriving in the New Data-Driven Healthcare Market.” If you like the show, please do us a favor and leave a rating and review at Apple Podcasts.
Harry Glorikian: My guest today is Matthew Might. He’s a computer scientist who transitioned into precision medicine and now builds computational tools to find new treatments for rare diseases. Since 2017 he’s been the director of the Hugh Kaul Precision Medicine Institute at the University of Alabama at Birmingham.
Might’s journey from pure computer science into medicine is a deeply personal story that began with the birth of his first child Bertrand in 2007. Bertrand seemed healthy at first. But soon he showed a mysterious cluster of symptoms including seizures, lack of motor control, and inability to produce tears.
For more than four years, doctors were unable to diagnose Bertrand’s condition. But eventually, using a then-new technology called whole exome sequencing, they determined that he had no functioning gene for NGLY1, an enzyme that normally helps to clear junk proteins out of cells.
It turned out that Bertrand was the first person ever to be diagnosed with NGLY1 deficiency. There was no obvious treatment available. Matthew says that’s when he began his transformation into an amateur biologist.
He shared Bertrand’s story on social media and in the press, and was able to locate and organize the families of dozens of other patients around the world who had the same mutation. He worked with colleagues at the University of Utah to make genetically engineered planarian worms that had a similar mutation. Experiments on the worms led showed that knocking out a second gene, for another enzyme called ENGase, seemed to help the worms live longer.
So on a hunch, Might set off on a computational search for compounds that might bind to ENGase in humans and inhibit its activity. He discovered that there was a drug on the market called Prevacid that was approved to treat acid reflux but also, as an unexpected side effect, inhibits ENGase. So Bertrand started taking Prevacid, and it helped. Matthew says it was one of the treatments that helped to extend and enrich his life.
Sadly, Bertrand passed away last year at the age of 12. But by that point, his father’s work to apply computation to basic biology, and thereby speed up the treatment of rare disorders, had sparked a movement that will long outlive him.
The story caught the attention of the Obama White House, which asked Matthew to lead several new initiatives in genomics and precision medicine. One of those was a pilot called the Patient-Empowered Precision Medicine Alliance, which had the goal of quickly diagnosing rare conditions and identifying treatments for more patients.
Now Might is continuing that work at the University of Alabama, Birmingham, where the Precision Medicine Institute uses computer science techniques like knowledge graphs and natural language processing to find more drugs that can be repurposed to fight rare diseases.
We covered all of that ground and more when we talked in late August. And once you hear our interview, I think you’ll agree with Might that computation is accelerating the genomics revolution in a way that’s going to change healthcare not just for people with rare diseases, but for everyone.
Harry Glorikian: Matthew, welcome to the show.
Matthew Might: Oh, thanks. Good to be here.
Harry Glorikian: [I] spent a lot of time reading about what you’re doing your past and sort of the history here, but I want to start off with, which sort of which fits right into the show, is you’ve said that data is the greatest drug of the 21st century and that precision medicine delivers data as a drug. Can you expand on what you mean by that
Matthew Might: Yeah. And I’ve said this a few times in a few ways, but the principle here is that I think we need to look at data itself as a kind of intervention,that exposure to one’s own data could have ramifications for your health. And you can imagine this in a very general sense.
Like if you get detailed data about your health and you might do something about it, but if you give extremely detailed data to your physician, they might be able to do something with. and, and oftentimes I’m thinking in terms of the very molecular in that case. And that’s really where I spend a lot of my time, but giving clinicians molecular resolution on the nature of your specific health problems is really what I think is so revolutionary about medicine, right now. And the ways in which we can gather that data and the computational tools that will be available someday to physicians, and even to a certain extent right now will enable them to do things that no drug can do on its own.
Harry Glorikian: Well, it’s, it’s interesting that you say data and, in that sense is I was just playing with something that pulled in all my medical data and put out all the longitudinal charts for me and highlighted all the ones where I was out of whack. And I texted my doctor. I’m like, we need to get on a Zoom call. I need to show you a couple of things that are out of whack, and I want to figure out why they’re out of whack. So I agree with you that molecular will be that, that really high resolution level to get to. But even I think the simple data to give patients I think is powerful. It can move the needle, if we can communicate it to them effectively.
Matthew Might: Yeah. I mean, I think even the simplest incarnation of this could do some good, like imagine if your scale not only told you your current weight, but just drew a line between your last three readings and said, this is where you’re going to be in 10 weeks. Something that simple might help.
Harry Glorikian: My scale does do that and it tells me that you are getting fat, so you gotta, need to do something. So I try to intervene when I can. But, so, you have an interesting history and past, I mean, you went from pure computer scientist from the university of Utah into precision medicine at the University of Alabama. I mean, that’s, I almost want to say based on what I was reading like that revolves around your personal experience from finding a diagnosis and treatment for your son. Can you give the listeners may be a condensed version of that story and how did it turn out that your experience studying things like functional programming turned out to be so useful for studying rare diseases?
Matthew Might: Yeah, well, that’s a wonderfully broad question. And yeah, I’ve had an unusual path to this point. So I’m currently the director of the Precision Medicine Institute at UAB. And so it’s, it’s a, very focused medical research institute. We want to help patients find tailored therapies for them.
My background in computation and computer science certainly influences that. We have a host of computational tools to help do that. Some are based on artificial intelligence. Some are very systems biology focused, that start to invoke aspects of functional programming. And you’re right. And the reason I started all this is my oldest son, who unfortunately passed away in October, had an undiagnosed [disease]. And for four years I had no idea what he had. Eventually through a novel application of exome sequencing was able to determine that he had the first case ever known of this ultra rare disease called NGLY1 deficiency.
And I think that’s safely the point where I really flipped in my head from a computer scientist to an amateur biologist. I knew enough to try to get him diagnosed, but that’s one where I said, I’ve got to find some way to help him. And even though he passed away, it’s hard to look back on his life and see it as anything but a major success in many ways, because he was born with a very short life expectancy and yet he made it to almost 13 years old and that would not have happened without sort of a sequence of emerging technologies that came just barely in time to extend and enrich his life and bring him a lot of joy.
I miss him every day, but I’m lucky that I have the opportunity to work every day, literally every day towards his legacy of helping patients with science. A lot of which is computational, but much of which is sometimes just good old fashioned wet biology where we go to the bench and try something out.
Harry Glorikian: Yeah. I mean, you’ve sort of described the Precision Medicine Institute as a form of research consultative service, where the goal is to find the next step for any patient that reaches out on their diagnostic or therapeutic odyssey. I mean, that sounds amazing. I mean, maybe you could describe more of what happens on a day-to-day basis.
Matthew Might: Yeah. So, it all comes down to Monday. So Monday is case review day. So if someone has reached out to the institute or if we are sort of currently working on something for them, Monday is where we all synchronize and put our heads together and try to figure out that next step.
So for patients that have reached out for the first time, it’s, “Okay. What’s the direction of a therapy.” And for those that are in flight, if there’s been a change, if some experiment has completed, if some lab has come back, if new information has been introduced, we check to see, is there a new next step? Is there some, is there some new insight? And sometimes on those Mondays information will come back that enables a query to be run on one of our computational tools where, the best example would be, targets emerged. Like, if we modulate the behavior of this gene, we think it will be therapeutic for this patient now. And so we can run queries to see, can we modulate the behavior of that gene using some kind of small molecule or some other approach.
Harry Glorikian: So, but when you’re doing all this, I mean, are you, do you have a mission to sort of either scale up or automate? And if you do, how’s that going?
Matthew Might: Yeah, so it’s, it’s both. And in fact, it’s, it’s scaling through automating. We realized pretty early on that humans are an essential part of this process, right now, in the sense that humans really do need to—in this case, when I say humans, I mean, undergraduate students, because they’re the ones within the institute that act effectively as the case managers and reach out and sort of pull in the information, digest it to some kind of structured format that the tools can process.
They might engage with the physician. They might reach out to some basic scientists that have insight on the relevant biological processes and figure out how do we drive it to a query or a recommendation for an experiment. And so, in some sense, when I think about scale, what I’m really thinking about is the efficiency of these undergraduates. How many cases per week can we get them to process and how much tooling and automation can we build to make them better and better at what they do? So that’s how I think about scale.
And then I think about replicating this as a process and every academic medicine, medical center across the country. There’s no reason you couldn’t have a team of extremely bright students in every center of the country that run this kind of process for their patients locally.
Harry Glorikian: Yeah. I mean, I would think that that would be one hell of an experience for the students to sort of see something actually being practically applied,as opposed to reading it in a book and it being a little bit more theoretical.
Matthew Might: It is actually, in fact, I’ve noticed that literally 100% of the students that have participated in this program, I mean, all of them have gone on to graduate school, either for an MD, a PhD or both. So it’s a 100% success rate to getting students in the grad school or med school.
And now we have a course version of it. And so in fact, several course versions of this, where you can take this class and you’ll get to practice on some existing solved cases, but we even throw some unsolved ones in the mix to see how they do. And when they take these courses, and then for the honors students now at UAB, they can take a special course sequence as freshmen, where they’ll get into the lab and build model organisms that represent some of the patients, which could ultimately enable the discovery of therapies for them. So I have to say it’s, I can’t think of a lot of other places where you get that kind of experience as an undergraduate.
Harry Glorikian: Well, no, that’s what I was thinking. I was thinking, “Hmm. How do we get this more broadly out there so that more people are doing this and, and get their head in that, in that zone and understand these issues.”
But I think one of the projects that I was looking at was mediKanren, if I’m pronouncing it correctly. What is it meant to do?
Matthew Might: So mediKanren is really our flagship artificial intelligence tool that we use primarily for drug repurposing. We kind of built it with the end application in mind. So I’ll tell you what it’s really good at doing. If you tell it a gene and you tell it a direction to go, like, I want to make this gene more active or make this gene less active, it does very well on those kinds of questions and it can scour a number of data sets to do this.
So we’re part of a, actually a larger effort through the NIH called the Translator Consortium. This is a huge research effort. We have lots of teams working together to both mine out all biomedical knowledge and make it structured. And another set of teams are trying to do automated reasoning on top of all of that knowledge. So we’re on the automated reasoning side. We can do some of the mining too, but, the other teams do such a fantastic job that we mostly just consume what they produce in terms the mining. And then we try to stitch it together, so that we can find interesting ways to go after targets of interest.
Harry Glorikian: So it’s reasoning over medical knowledge graphs, I think, that you’re trying to do. And so it sounds like a promising way to find unexpected connections between diseases and existing drugs. But if you had to explain that to a layman, how would you explain sort of a knowledge graph and what you guys are doing with it? Or do you have a favorite example?
Matthew Might: I have pictures, but I can also do it with words. Knowledge graph. So let’s talk about what it represents. Ultimately, there’s a structure, but that’s not actually all that important. A knowledge graph is a collection of facts, and in facts are sentences. And they’re sentences of the form “A somehow relates to B in some sense.” and knowledge graph is just a huge collection of these sentences, “A is related to B,” where there’s a specific relationship.
So a biomedical knowledge graph is going to have some constraints on it. So the A’s and the B’s that you’re connecting are going to be nouns from medicine and biology. So there’ll be drugs and diseases and genes and metabolites and all the other stuff that you typically read about in and medical papers. And the relationships are going to be biomedical in nature too. So it’s going to be things like A inhibits B, A activates B, A causes B, A treats B. And so if you collect all of these sentences together, you have what we call a knowledge graph. And the cool thing about a knowledge graph is that you can do logic on top of it and try to look for relationships that are there, but not explicitly stated.
The simplest example of this is let’s suppose there’s two sentences in this knowledge graph, there’s Aa increases B and B increases C. We can infer logically that if you increase A, you should also increase C, because B went up and so C should go up. So that’s an example of logical inference on top of one of these knowledge graphs.
Harry Glorikian: And so that’s typically—there’s a human intervention at some point to sort of look at this and then say, yes, this makes sense?
Matthew Might: Yes, absolutely. So one of the major roles of these undergraduate analysts is to actually double check what comes back from a tool like this, because it’s going to admit a logic argument. It’s going to say, “I believe that this is going to influence the right target because,” and then the analyst can look at the because, and it’s going to have references into biomedical data sets. It’s going to have references to papers in PubMed, and they can go read those. They can look at the data sets and they can double-check the reasoner and say, you know what, I think you got this right. Or no, you made a mistake. And it does make mistakes sometimes. So a lot of the knowledge from the literature has been done by natural language processing and that makes mistakes. It’s critical to have a human in the loop to double-check that.
And towards your earlier question about how we do scale, one of the things that we’ve added to the tool is ways to make that check go faster. So, for example, on the latest interface, when it tells you that it believes, for example, A increases B and you click on that, it’s going to jump straight to the sentence that it pulled that from in the paper. And so you can just look directly at that sentence and say, do I believe this? Do I believe it got A right? Do I think it got B right? Do I think it got the relationship right? It’s sitting right in front of the analyst. Whereas previously that was a few clicks away. They had to click on that. They had to click on the paper it found they had to go to the paper on their web browser. They had to look at the abstract, they had to find the sentence that it got it from, and then figure it out. That’s a long process actually now, and going from, a few minutes to verify and inference to a few seconds, that’s a huge increase in efficiency for these analysts.
Harry Glorikian: one of the things I would say is I always try to find out, is it shortens the overall process of even finding this relationship. I mean, if you had to put sort of time scales on this, how much faster you think that we’re speeding up this whole process of being able to even identify something that might have this effect?
Matthew Might: Yeah. Yeah. I mean, we we’ve had some natural experiments in this regard where in some cases there were answers sort of buried in the literature that seemed to have been therapeutically relevant and yet, very motivated disease communities hadn’t stumbled across them, and motivated physician-scientists researching these diseases had not run across them, or didn’t sort of connect the dots to realize that this could actually be relevant to a patient. Probably the most recent example of this is ADNP-driven autism, where there were results in the literature that could imply the key finding, which is that low-dose ketamine will increase ADNP. That’s the key thing that the researchers trying to treat this disease were after. And in some sense that was out publicly known, if you will, in the literature. And yet it took running this query to find it,to sort of make the realization that this was true.
It’s kind of interesting actually to think about the fact that there’s stuff out there that as a species we know, but we don’t know that we know. So we call that sometimes the “unknown known.” It actually happens a lot in different contexts. And I even remember this happening in computer science, where there were communities out there so disparate that one had solved a problem the other had been trying to solve for three decades, and they just didn’t know that I had effectively been solved. I mean, it can happen, actually.
Harry Glorikian: Yes. And I talked to different groups that are working on systems that will make those unknowns more easily findable, or at least highlight them so that people know they’re there. But you guys search scientific literature, drug databases, for existing and approved [drugs]. And basically you’re looking to find something that’s going to perturb an ultra-rare disease. Why is it better to look for an existing drug rather than a new one? I’m just, curious of the practical arguments around that.
Matthew Might: I’m not against developing a novel drug for a single patient. It’s just that most patients don’t have $2.6 billion. So it’s a little out of their price range. That said, of course there’s technologies that are changing this equation substantially. So I would say oligonucleotide therapies in general, it’s not down to a thousand dollars a patient, but it’s dramatically less than $2.6 billion. We’re probably closer to the range of a couple million dollars, and that’s falling fast, to do these sort of custom-programmed therapeutics for individual patients. So, yeah, I’m not against finding novel matter. It’s just that it’s still outside the budget of what most ultra-rare patients can handle.
Harry Glorikian: Right. Right. Well, I was talking to, it hasn’t even come out yet, but I was talking to Kevin Davies in one of my last podcasts about CRISPR and just exactly that same discussion. So I don’t know if you guys have sort of done a ballpark or sort of a thumbnail of what do you think, what fraction of rare conditions do you think, are treatable the way that you’re you guys are approaching it? Is it fair to say that eventually you’ll exhaust that approach in that we’ll have to develop a new drug for the next disease?
Matthew Might: Yeah and I guess we’ve got to sort of clarify what we mean by approach. So there’s the AI-based approach or sort of strictly computational approach. And then there is what we can do if we’re allowed to go to the wet lab for a little bit of stuff.
If you play the game where you can only use a computer, there’s the answer for today and the answer in the limit. Once we’ve sort of saturated biomedical knowledge graphs, if you will, with everything wherever we’re going to know—and already, right off the bat, I think we jumped to a reasonable suggestion somewhere between 5 and 10% of the time, for the case of ultra-rare genetic diseases, and there are factors that can influence that. So for example, if it’s a dominant disorder where the genetic insult has really just sort of tweaked the thermostat on a gene, so it’s a little overactive or a little underactive, we tend to have a better success rate jumping straight to an answer with a computational tool than if the gene has been wiped out and now we have to find a way to replace that activity.
Now the good news is if you look at sort of the census of rare disease, 70 or 75% of all patients fall into that bucket of the genes have become a little overactive or a little underactive. And so it’s very amenable to an approach like this. And for the patients that where the gene is missing, there are still things we can do computationally. The call I had right before this one was exactly that case for you. What can we do computationally? And by playing with the tool, we found some alternate targets to go after. But it takes some play to do it at that point. It’s not quite as automated, but you’re still using the tool as targets emerge to ask the right questions.
Harry Glorikian: I want to pause the conversation for a minute to make a quick request.
If you’re a fan of MoneyBall Medicine, you know that we’ve published dozens of interviews with leading scientists and entrepreneurs exploring the boundaries of data-driven healthcare and research. And you can listen to all of those episodes for free at Apple Podcasts, or at my website glorikian.com, or wherever you get your podcasts.
There’s one small thing you can do in return, and that’s to leave a rating and a review of the show on Apple Podcasts. It’s one of the best ways to help other listeners find and follow the show.
If you’ve never posted a review or a rating, it’s easy. All you have to do is open the Apple Podcasts app on your smartphone, search for MoneyBall Medicine, and scroll down to the Ratings & Reviews section. Tap the stars to rate the show, and then tap the link that says Write a Review to leave your comments. It’ll only take a minute, but it’ll help us out immensely. Thank you!
And now back to the show.
Harry Glorikian: Before you moved to UAB, you were working on precision medicine initiatives for the Obama White House, I believe, and then briefly for the Trump White House. Can you update the status of the, was it, the All for Us initiative and then the Patient Empowered Precision Medicine Alliance?
Matthew Might: Sure. So, yeah, I did spend, I think, in total, about three years either working with or working for the White House, under both Obama and Trumpm on the precision medicine initiative and related initiatives, like the Million Veterans program, as well as micro-initiatives, like the Patient Empowerment Precision Medicine Alliance, where we were just kind of trying to test it out, to see if these ultra-tailored approaches would work.
So as far as where things stand today, All of Us is a very successful, large scale clinical genomics research program, which is, I think on the way to hitting its target of enrolling a million Americans. And the way I described that program even at the time was like you’re trying to build the Rosetta stone of the human genome. It turns out we reached a point where it’s really easy to sequence a genome. Not so easy to interpret a genome. So if I sequence you or me and we find mutations, we go, “huh. Well, that’s interesting. What does that mean?” And we go, I don’t know. But let’s suppose you got a million people that donate their health records and their genomes. Well, now you can start to draw statistical connections between what this mutation or this collection of mutations means in terms of actual human health. So it’s finally a way to start decoding it. And so that’s really what All of Us is about in my view. It’s about building that Rosetta stone for the human genome.
For the Million Veterans program, it’s actually sort of inside of it, yet also outside of it and really trying to do the same thing, but leveraging the extensive clinical records and histories that the VA happens to have access to, and taking a slightly different method to get there in terms of genomic data, starting with SNP chips and genotyping, as opposed to full-on sequencing.
And then for PEPMA, that was really a pilot project where we just said, okay, can we take a small number of patients and can we actually sort of run this process all the way forward, where we get a genotype and try to find a medication that might help them. And it turns out for a handful of cases, we were actually able to do it. It was all thanks to getting a group of private entities, like, companies and universities to come together so that we actually had enough infrastructure in one place to run the process. And so we were actually able to do that.
Harry Glorikian: Yeah, it’s funny because I always think to myself, like all these silos, if we could just have them integrate in some way, you’d have a lot more data to work with. I always find that in the beginning, you always find those low hanging fruit that sort of fall out and then it gets harder. If you have enough to start with, something interesting falls out of it.
You’ve said that in precision medicine, for a lot of cases that we deal with, we don’t have sort of the right drug right away, but we can always prescribe an experiment. What do you mean by that?
Matthew Might: Yeah, by that, I mean, and so this gets into sort of like the philosophy of medicine itself. So you’ll hear clinicians use terms like “This is not actionable.” And then you hear that an awful lot in rare disease. You hear it a lot at the end of cancer, where they’ll say, oh, there’s nothing we can do, or there’s no clinical utility in this. And, and I think precision medicine subverts that whole approach and says that, well, if you’ve run out of clinical options, you can still do some science.
And increasingly I think we can systematize that science, so that it’s not okay. We need to do something. It’s, here’s a set of things that you could reasonably do at this point, a set of experiments that if run, might point to what to do or might point to another next step. And a great example of that is, for, for, particularly for a rare genetic disorder, in many cases you can build a model organism, you can build a fly and you can build a worm pretty inexpensively, to model that genetic disease, using things like CRISPR. People don’t appreciate, I think, the full importance of gene editing. People think about editing human genomes, but the fact that it made it so much easier to edit animal genomes was actually in many ways an even bigger impact and a more immediate one as well.
So yeah, you could for virtually any genetic disease, if there’s an animal equivalent to that gene, you can build the animal. And if it’s a small animal, you can test a lot of drugs on it, pretty inexpensively. So, I’ve got a friend and collaborator, Ethan Perlstein, who built a company around this approach and was very successful actually in treating, some patients this way. I have a collaborator, Clement Chow, who is an academic doing this on the academic side, focused on Drosophila, doing this very successfully.
So, that’s not a drug, it’s not a procedure, not a medical procedure anyway, but it’s a very well-defined process. And it’s a process whose outcomes could be measured statistically. So you might even know what your odds of success are before you try it, whether or not you’re going to find something. So I think it takes an evolution in our thinking to realize that this is a perfectly reasonable thing to do for a lot of patients: build that fly, build that worm and test a bunch of drugs.
Harry Glorikian: And, there’s a lot of times where things seem perfectly normal for me to suggest, and I’ve had people look at me, like I just grew like two more heads off, off my shoulders.
So, but this sounds like if this is your fundamental belief that there is nothing that is not actionable, medicine or otherwise.
Matthew Might: Right, right. And I think it also requires a degree of stoicism because just because there’s something to do, it doesn’t mean it’s going to work in time. And this was something I was always mindful of during my son’s life, was that while there was always something to do, I was mindful that it was probably not going to always happen in time. It was always a race against the clock. But there was always something to do. And even today there’s still, as I say, I’m still working on his condition. I’m still very actively engaged in drug development for his disorder. So even now there’s something to do. As a parent it’s still brings me benefit to know that it, it will benefit others. So it does require a shift in perspective about the meaning of actionable.
Harry Glorikian: It feels like finding [computational] ways to use existing drugs, to help people with rare problems, was waiting to be solved with someone with your exact skillset in computer science and your exact set of motivations as a father of a child with a genetic disorder. And so many other key players that I’ve talked to in this sort of that have this N of 1 stories have very similar biographies. I mean, there’s been a few movies made about it, right? It makes somebody wonder that if you know your son hadn’t been born to you with your expertise, who could apply knowledge and bring that experience to it, that you wouldn’t be moving the ball forward. How does that make you feel about the state of science or medicine?
Matthew Might: Yeah. And, you’re not the first to make an observation that I sort of ended up in the right place at the right time with the right set of motivations. And there’s a lot of truth to that. I think about if he’d been born even a few years earlier, or a few years later, how things would have played out differently.
I realized early on that there’s a desperate need for computer science within medicine, that there is so much opportunity just left on the table for lack of expertise. But I made a deeper observation, which is that as much as medicine needs computer science, what it needs even more is computer scientists.
The problem is that the average computer scientist doesn’t have sufficient motivation to go learn. Medicine’s big, it’s messy. And I got to say, biology is so messy that to the mind of a computer scientist at times, you’re just like, God, what a disaster biology is. It’s like every time you have a rule or a law, most of the time at the end of it, there’s nothing ever that’s always true. And when you come from a field like computer science, where you can put clean little theorems around everything and layers of abstraction that never break, it’s like, oh gosh, who designed all this? Who was the engineer that thought that was a good idea?
That that’s how I feel half the time when I dive into biology. And yet there are abstractions that you can borrow from computer science and you can use these things to start to describe the way biology does what it does. And so I do think of the cell as a computer or a machine—maybe a Rube Goldberg machine, but a machine nonetheless. And one that you can sort of intellectually approach, from the direction of computer science.
Within computer science, I happen to have a background in functional programming. And there are times when, describing the mechanics of how biological processes operate, where this kind of feels like, I’m playing with a little functional programming language. Like I’m doing graphic writing instead of term rewriting. There’s been these moments where I’m like, yeah, this is just a programming language. It’s weird, but it is one. And then I think, gosh, I mean, it is strange that I arrived with that particular skill set at this point in biology’s history, to make these observations and use that profitably towards helping patients.
So in terms of how they makes me feel? Lucky, I guess, that I’ve, I’ve just sort of been there on the right place at the right time. And the same thing is true for almost everything else that’s occurred since, since Bertrand was born, from the timing of the precision medicine initiative itself, to getting his story in front of President Obama at just that moment, to getting the invitation to go, to getting the invitation to participate, and then join the team. I mean, the timing on all of it was just so ridiculous to me that I look back and think, I can’t believe that happened.
Harry Glorikian: Wow. I mean, timing. Being in the right place at the right time, a little luck, I’ll take that every day, right, where everything starts ti come together.I think back to, because I was involved at Applied Biosystems when we did the genome and wow, that was such a big deal. And then every once in a while I still see an article saying. Yeah, the genome hasn’t really done much. And I’m like, these people, how do they write these things with a straight face? And it gets published in a reasonable journal. And I’m like, these people are out of their minds, considering everything in biotech, everything in, functional genomics, all this stuff is, is grounded in that information.
Matthew Might: I see the same stuff and I think, what do you mean nothing has come out of it. What are you talking about? Everything has come from that. And then, when I point to success stories with individual patients, which are growing and growing, they’re like, yeah, but that’s the exception. It’s turning into the rule more and more, and I think what you’re seeing now is that as with any new technology, the barrier to entry starts very high. But that barrier has been falling fast to the point where, people who start off, the parent side, like me, are increasingly finding that they can get into the game and that they can do something.
And I think it’s at a level now where almost any, patient or parent that has a technical background can jump over and do something. But even patients without that background are making the jump now, too. So barrier to entry is falling so fast that it really has changed everything when it comes to patients moving the needle for themselves using the fruits of the genomic era.
Harry Glorikian: Yeah. And I think computational, power and costs and ease of use are starting to come down dramatically, which then brings the two together, which is of course the idea of behind the whole show and everybody that I talk to, and I see the, some of the companies I talked to they’re like, yep, we sort of eliminated three years of work. We could get it done in, a week to two weeks because of what we’re looking at, how we’ve applied our computer science. How many new pathways we can sort of identify of course, for new drugs.
Matthew Might: And I, I can give you examples of where the barriers fell overnight as I needed them to, just by luck. Or when it came to creating model organisms, right? Before CRISPR, gosh, that was an expensive, daunting. process, it took a lot of time. And then CRISPR shows up and they’re like, oh yeah, no, it’ll be a few months and $10,000. And it was just, I mean, just like that it happened. And there’s equivalent revolutions happening on the computational side too. If you look at your protein folding technology, it was a joke, that like, yeah, it’ll never happen in silico. And then all of a sudden, like now some say maybe the only way we’ll ever get some structures is in silico. And then that was kind of an overnight thing too. Obviously it wasn’t overnight for the engineers on the project at Google. But once it appears like, oh my gosh, what a game changer.
Harry Glorikian: Well, and then as soon as somebody does it, it motivates more people to sort of grow and it sort of moves the space forward that much faster. That’s the part I find interesting is most people have trouble understanding the speed of change, and it’s moving faster now than—and I’m used to, trying to keep track of how fast everything’s going, and I’m finding myself having trouble keep up with how quickly things are shifting.
Matthew Might: It really is changing faster than I think any one person can predict. And the disruptions are coming almost out of nowhere. Like no one saw CRISPR coming. You might reasonably foresee that at some point, some efficient gene editing technology would have emerged. But I think it emerged much faster than was expected.I remember when I would work with patients, five or six years ago, I’d say, yeah, there’s this thing, these antisense oligonucleotides, and maybe someday, but we’re probably, I would say at the time, like maybe 20 years away. Then you see oligonucleotide therapies really take off and, then I think it was two years later there’s an FDA approval. Then a couple of years after that, there’s the first big N of 1 introduction. And then like a year and a half later, we were all injecting mRNA into ourselves. Well, that happened pretty fast. It wasn’t a couple years.
Harry Glorikian: Yeah. And, and for people like you and me that are in this, like, my, my mind is like, wow, this is awesome. And then I try to explain to someone and they don’t understand the impact that some of this is happening in the implications of what we’re talking about.
Matthew Might: Yeah. And, I think that, going forward, it’s going to be a much steeper acceleration than anybody can really predict because we’ve suddenly just burst into the era of programmable therapeutics. I mean, COVID really suddenly just threw it on the table. There it is. And an example as well, people said, okay, well, if you can just give mRNA directly, instead of trying to deliver these complicated proteins to do the gene editing, why don’t you deliver the mRNA for the CRISPR protein or, for, for CAS9 and deliver this along with the guide RNAs, well that’s much easier. And my, my gosh, it looks like it might actually work. So these things, they couple in unexpected ways, and very quickly too. And so I I’m excited cause I have no way to know what’s coming now.
Harry Glorikian: I’ve always felt, I don’t know what’s coming. That’s why I try to read such a broad array of, sources, everything that’s going on in, you know, chip development to what’s going on in our world. But I think the next big wave of shifts is going to be how a lot of this gets implemented, the business models behind it. And that’s the next big shift because you don’t have to do it exactly the same way you had been doing it up til now.
Matthew Might: Oh, I agree. And, and oddly enough, yeah, I spent a fair amount of time thinking about stuff as mundane as how do we get payers to actually pay for some of these things? How do we show them that there is value to be captured already? And, because there is, I think we’re not far away from a future where payers realize that it’s going to be cheaper to take this very expensive patient with a complex disease and look for sort of a root cause treatment than to continue paying for symptomatic treatment. I think we’re at the threshold of that era.
Harry Glorikian: Well, I think, the CEO of Illumina said we want to get whole genome down to $60. Right. I mean, at some point you’re like, okay, when are you going to stop being worried about the cost of this? Because it’s going to be a rounding error at some point.
Matthew Might: Yeah. Over the course of someone’s life, it’s already a rounding error you know it’s already there.
Harry Glorikian: But $60, yeah. I mean, I was, I was, talking to a company where they could do, if you could do the initial analytics for $60 and then do the computational on top of it for another $60, at some point you’re like, look, we should just be doing this for everybody. The problem is the implementation. And can physicians keep up with, what does it all mean and what am I supposed to do?
Matthew Might: Yeah. And that’s why I think, when I talk about precision medicine and data as a drug, I always have to highlight the importance of computational aid for the physician. Because if you were to give a physician [raw DNA data[, they would go, “What, I don’t know what to do with that.” Even if you distill it down to the individual mutations, the average physician goes, “I still don’t know what to do with that.” It’s gotta be broken down into something far more actionable for them.
And I think we’re going to look back at now as sort of like the dark ages of IT in medicine, because we’re in a situation where I don’t know any physician that loves the EHR they use. In fact, they all hate it. It is a disastrous user experience across the board. And this is a classic problem in software where the people who pay for the things are not the people who use the things, and say, so what are EHRs optimized for? Billing. There’s only one EHR as far as I can tell it’s optimized for patient care, and that’s at the VA, where they’re not really concerned about billing. And so people like that one, which is, not, not a big surprise.
Harry Glorikian: Well, and they were talking about, they wanted to put in Epic. I was like, who got paid like behind some closed door to make that decision? That was the dumbest decision I’ve ever heard anybody make.
Matthew Might: I thought the same thing as you, having worked in the Million Veterans program. Like, no, that’s the crown jewel. That thing actually works and it works well, and it gets great data, do not replace that. Keep it as is.
Harry Glorikian: Yeah. Well you need to, unfortunately whoever’s making that decision has no skin in that game as far as I can tell, but I agree with you. I mean, I’ve said over and over, if anything’s gonna break medicine, it’s going to be the existing EMR systems because you can’t innovate if you can’t get the data out. And Google and Microsoft and Apple and everybody’s innovating because they get to change their system at will, right. Everybody gets to jump on AWS and innovate. The system is sort of stuck in stasis and can’t move out of it, which is what I find worrisome.
Matthew Might: I agree. You’ve either got to get the data out or the computation in. Preferably both. I’ve dealt with physicians where I can say, “Hey, we could give you this really cool genomic test for your patients. And then if they try to take a drug, you’ll know if it’s not going to work for them.” And they go, “Well, will there be automated decision support in the EHR to tell me if that happens? Or do I have to sort of look at the note and see that they have this variant?” I go, “Well, you can have to look at the note.” And they say, “No, I do not want that, because if that note is in there and I don’t figure that out, and I prescribe a drug that causes an adverse event. I’ll get sued. But if the information’s not there at all, I can’t be sued.” That’s the world we live in.
Harry Glorikian: Well, listen, it was great to speak to you. The stuff you’re doing is awesome. I wish more people knew about it. I wish more students were involved in it so they could get firsthand experience. Like you said, I think that’s when we can start to teach people the crossover between medicine and computational work in general, because I’m always trying to find people that know both, and there’s not a lot of fruit on that tree at the moment. More is growing, but not as much as you’d like.
Matthew Might: I agree. We need to get people going more often in both directions. And that’s one of my missions at the Institute as well as to cross-train folks in into both sides, biology and computer science.
Harry Glorikian: Excellent. Well, it was great to talk to you. I appreciate the time.
Matthew Might: Likewise. It has been a pleasure.
Harry Glorikian: That’s it for this week’s show. You can find past episodes of MoneyBall Medicine at my website, glorikian.com, under the tab “Podcast.” And you can follow me on Twitter at hglorikian. Thanks for listening, and we’ll be back soon with our next interview.