Dr. Frank Maddux: The escalating incidence of infections caused by drug resistant bacteria contributes to a reported 700,000 deaths annually. That number is expected to soar to over 10 million by 2050. Timely, actionable information is necessary to provide effective treatments quickly to drug resistant infections. Today's guest, Dr. Miriam Huntley, chief technology officer for Day Zero Diagnostics, talks with us about their efforts to improve infectious disease diagnosis and treatment with genomic sequencing and machine learning to solve antibiotic resistance.
Dr. Miriam Huntley: Thank you for having me.
Dr. Frank Maddux: Tell us just a little bit about Day Zero and sort of what is it you all are all about?
Dr. Miriam Huntley: So at Day Zero Diagnostics, we're developing a new diagnostic for bacterial infections. Essentially the way bacterial infections are diagnosed today is using a slow, outdated method. And we're trying to develop a new diagnostic that uses whole genome sequencing and machine learning to develop a faster, better diagnostic.
Dr. Frank Maddux: In my practice over the years, we would we would pull a series of blood cultures. We would get them plated around a lab or with a reference lab or a hospital and it would take usually two days, three days to get a preliminary result. And Then the actual antibiotic resistance results might come back even later than that. What is it that you all actually do to try to reduce this time?
Dr. Miriam Huntley: Unfortunately, the way that actual infections are diagnosed today is using culture and growth of bacteria on plates or in bottles. They can work really well. But as you mentioned, it can be very slow. And so By the time the results comes back, the patient can have already decompensated, and it makes it very difficult for the doctor to prescribe the right antibiotics. So our approach is instead of waiting for bacteria to grow, which is slow, we try to diagnose directly from the clinical sample, so directly from the blood sample or other clinical sample. To do that, we first sequenced the pathogens genome that's found in the sample. Then we use whole genome sequencing of the bacterial pathogen. Then we use machine learning in order to figure out how to use that information to provide clinical diagnostic information. So, to determine what the species of the infection is and what the antibiotic resistance is.
Dr. Frank Maddux: My first question would be how do you differentiate bacterial DNA from human DNA? Because these samples surely have both.
Dr. Miriam Huntley: It turns out that in a bacterial infection, so a patient who is infected with a with a bacteria, let's say in their blood, they could be dying and bacteremia could be extremely dire. The amount of human DNA in that sample far outnumbers the amount of bacterial DNA. There's about ten to the eight more human DNA in these samples than bacterial DNA. So, eight orders of magnitude more human DNA. If you just try to sequence everything that’s in that sample, if you try to sequence all of the DNA, the signal-to-noise is too tiny, you'll never actually see the bacterial DNA. And as you said, you'll never be able to distinguish it from the human DNA.
Our approach at Day Zero is instead to use a technology that we developed to actually enrich the bacterial DNA out of this sample. If you can think about it as an in the sample, the bacterial DNA is a needle in a haystack where the haystack is the human DNA, and the bacteria represent the needle and traditional approaches to this problem. Using PCR based probes, they try to pull out the needle, and they have to kind of develop really specific probes to find the bacteria that they're interested in. Unfortunately, there's only so many probes that you can develop with this approach You're not really sure if your probe comes up with nothing if you just don't have the right probe. Our approach is quite different from that. The technology that we developed first gets rid of the haystack which is a little bit counterintuitive. What that means is we're allowed to look at whatever's left, and we're able to sequence in an agnostic manner whatever bacterial pathogen is left over after we've burned the haystack.
Dr. Frank Maddux: If I have a patient that's infected, and let's take a peritoneal dialysis patient where I'm worried they've got peritonitis, and so we're going to look at their fluid. I'm as concerned about a fungal infection as a bacterial infection. And I'm concerned about multiple bacteria. How do you distinguish all these different types of infections and species?
Dr. Miriam Huntley: There's hundreds of pathogens that infect humans and hundreds of bacterial pathogens as well. And it's really important to be able to distinguish that long tail if you don't have the top five E coli, pseudomonas, enterococcus, staph aureus, specifically for a lot of the patients in this cohort, then you still want to know if they have anything else. It's really important to be agnostic, to be able to see whatever's there. Our diagnostic is specifically tailored right now to bacteria. So we're not working with fungi and viruses yet, though we are developing a platform that we think is going to be broadly applicable to other pathogens once we get to those.
Dr. Frank Maddux: You've gotten samples; you've extracted the needle from the haystack. You began to identify there's a species. How do you figure out resistance patterns? How do you figure out that this E. coli is a special E. coli?
Dr. Miriam Huntley: Just having the genomic sequence is not enough. So genomic sequencing is great. It's a great new technology. It's being applied in a lot of areas. But just knowing the A, G, Cs and Ts of an E coli genome, nobody really cares about that. You're not going to give that to a clinician, and they're not going to find that interesting. So, the trick is how do you go from a genomic sequencing data point to useful information. In particular, what's the antibiotic resistance based on that genomic sequence? For us, our approach to this has been to use the new technologies that have really been developed and are really productionized in the last decade or so, which is using machine learning. We develop machine learning models that can take genomic sequence as input and then determine what antibiotics this particular pathogen that's infecting the patient is resistant to, and that allows the clinician to determine which drug to treat their patient with.
Dr. Frank Maddux: Wow. Just give me a sense, your sense of the degree to which drug resistance is a problem today and going to be a problem tomorrow.
Dr. Miriam Huntley: I think antibiotic resistance is unfortunately on the rise. I think that that comes as a surprise to nobody. In the US alone, there's 2.8 million antibiotic resistant infections per year. So, it's a growing problem. And long gone are the days that we can just treat everybody with penicillin and hope for the best. As new antibiotics are introduced, bacteria can become resistant to them as well. There's been studies that show by the year 2050, we're going to be an incredibly dire circumstances if new therapeutics and new diagnostics aren't introduced. It's a major problem for many different indications. Sepsis is one that we focus on and certainly for patients with kidney disease it's a it's a very big problem for those patients as well.
Dr. Frank Maddux: The Gates Foundation funded a company out in the Northwest that's looking at the genomics of viral viruses and phages and other things that actually impact bacteria. Is that going to affect your work?
Dr. Miriam Huntley: The question of are there enough antibiotics to actually treat bacteria? I mean, we're kind of running out of our arsenal. So we need new therapeutics, that's for sure. Whether those are molecular therapeutics, small molecules, traditional therapeutics, or they're a new phage therapeutics, we need a new arsenal, a new kind of drugs coming down the pike.
But for us as a diagnostic company, whatever the therapeutic is, whether it's kind of new phage therapy or new types of drugs, you need to have a diagnostic that can determine: is this therapeutic going to work on this particular pathogen? Because, again, pathogens, no matter what the drug is, they learn to develop resistance. In the same way that we need personalized medicine in cases like oncology, all cancers are not the same. No single one drug works across all cancers. In that same way, we need personalized medicine for infectious disease. We need to figure out what is the right drug to treat this particular infection that this particular patient has.
Dr. Frank Maddux: Let's think about the technology a little bit. So, we're in the middle of creating a very large kidney related genomic registry and our anticipation is as that grows, we'll sequence patients in large batches with traditional equipment, predominantly equipment that we've seen from Illumina and others. You all are using a new technology. You want to describe the Nanopore technology?
Dr. Miriam Huntley: Yeah, absolutely. So sequencing right? It's gone through a few revolutions over the past couple of decades where we're very advanced from where we were in 2001 with the Human Genome Project Publishing the first human genome. Right now, what people are calling third generation sequencing technologies is a new wave of technologies that provide slightly different capabilities than some of the more workhorse technologies like Illumina that are out there. For our application, we're developing a diagnostic where time is of the essence. A patient really needs to have the right antibiotic quickly. We need something that can provide a quick turnaround time. And one of the technologies that can do that in terms of sequencing is this new technology company called Oxford Nanopore. At this point, they're a few years old, not super new, but their technology is able to provide rapid sequencing so that we don't have to wait a day to get the sequencing results, but we can rather get the sequencing results after an hour so that we can provide the results to the clinician in a timely manner.
So, I think there's going to be a lot of very interesting new developments in the sequencing space. Depending on the application, whether it's kind of rapid turnaround time or whether it's high throughput, the accuracy that's needed or even some molecular characteristics that it's trying to capture. I think those different technologies will find slots in various applications in medicine.
Dr. Frank Maddux: Whether it's whole genome sequencing of bacteria, human, whatever, DNA, it's huge pieces of data. I mean, there's a lot of information here to try to process. So, to what degree has your product and the way you're developing your product required that everybody who's utilizing these tests is going to have to be in a very connected environment where not all the processing is happening on premises?
Dr. Miriam Huntley: When our device is sitting in a hospital, we anticipate the device sits in a clinical microbiology lab. So in a place where lots of samples get processed, similar to kind of the samples that you've sent for culture many, many times. But, after the sample preparation and after the sequencing, that data gets transported to the cloud. We anticipate doing all of our analysis not on premise, but in the large cloud system so that we can perform large scale analytics. We can easily scale up, compute as needed. I think there's a few diagnostic processes that are migrating in that direction, really harnessing the power of the cloud, harnessing power of on-demand GPUs or CPUs. So that we can really provide the precision that we need in a timely manner without requiring those devices to be on site, which can be challenging.
Dr. Frank Maddux: In our company in the United States, that part of our company, we probably pull about 30-to-35,000 cultures a year, and we have anywhere from eight-to-10,000 positive cultures per year. It’s a large amount. Today, it takes at least two and a half to three days minimum to get the key results. What should we be expecting for the future if this works?
Dr. Miriam Huntley: Our goal at Day Zero Diagnostics is to develop a diagnostic that can provide the right answer, give the patient the right drug on the same day that they walk into the hospital, on day zero. So, we're developing a product that will have an earned turnaround time of 8 hours. From the sample, so not from the blood culture, not from some downstream product, not some daisy chain downstream product of blood culture.
Dr. Miriam Huntley: But from the blood itself or from other sample types that we're working on as future indications to the final result, to species identification and to antibiotic resistance results have that be an eight-hour turnaround so we can actually treat the patient quickly and get them on the right antibiotic sooner rather than five days later.
Dr. Frank Maddux: Do you think there are epigenetic phenomena on these bacteria? Is it going to be hard? Will the bacteria learn to evade your identification?
Dr. Miriam Huntley: For species identification, that's an easier problem than antibiotic resistance determination. For species, I think there’s so much information in the genome that encodes species, that there's no way to evade that. For antibiotic resistance determination, there are some epigenetic mechanisms that our technology doesn't capture because we're only looking at the DNA. If there's things like methylation, we're not capturing that but oftentimes those epigenetic mechanisms are encoded by something else in the genome. So, methyl transfers might be encoded. Our machine learning algorithm may be able to pick on that. I think it remains to be seen if evolution of bacteria may be faster than our machine learning algorithm is able to pick up. But we anticipate, over time, our machine learning algorithm continue to learn so we can continue to add more and more data into our training data set and learning to pick up any new mechanisms that might arise.
Dr. Frank Maddux: As bacteria change and begin to evolve, you're going to have to have an ever-evolving library, basically. How fast do you think you can respond to adjusting the details of your ability to be sensitive and specific in the answers you give?
Dr. Miriam Huntley: We have a training data set, and that's really essential. Any machine learning algorithm needs a large training dataset in order to actually learn and provide useful diagnostic information. So, to date, we've collected and really put together one of the largest datasets of this type. This data set comprises of the whole genome sequences of pathogens and their antibiotic resistance profiles. We can use those examples for the machine learning algorithm to learn that association between genome sequence and resistance. We're constantly growing it, so we have one of the largest today, but it's also one of the fastest growing ones. We anticipate continuing to seek out more data, seek out new resistance mechanisms as they emerge so that we can keep up. I think one of the challenges in terms of keeping up with resistance will be making sure we get enough data into the algorithm and then follow on kind of FDA submissions to make sure that we can update the algorithm. You know, that the algorithms understand the differences, but we have to get that cleared through the FDA before that can be applied in the clinic.
Dr. Frank Maddux: So, the FDA process for you is a typical 510K process?
Dr. Miriam Huntley: That's right, It'll be actually a de novo for our first submission because there's not really any predicate for what we're doing. We're applying both genome sequencing and machine learning in a new space of infectious disease where there hasn't been a lot of that in the past. There's been certainly a lot of AI applied in kind of imaging in the diagnostic space, but in the genomics space and in the infectious disease space, it's new.
Dr. Frank Maddux: Are all of the specimens that you get, liquid specimens, are some of the specimens tissue? And the reason I ask is we have invested largely in a regenerative medicine tissue engineering company called Humacyte. And one of the things that appears real is that those vessels that they build actually are much more resistant to infection than PTFE and other materials that are used for vascular grafts. So, I'm curious, are you able to process anything beyond sort of blood, urine, and other fluids?
Dr. Miriam Huntley: We anticipate that the device that we’re building will be a platform for many different sample types. To date, we’ve done a lot of different fluid sample types, but we haven’t tried solid yet. We think that there might be some ways of swab or homogenization in order to get that through, but that kind of remains to be seen for the technology.
Dr. Frank Maddux: I’ve been very interested in how we as a company catalyze innovation and begin to translate that so that it becomes standard of care when it’s really good quicker. But you're the result of technology transfer out of academic institutions. You want to just describe Day Zeros kind of history, and where did you come from? Who got this started? And why and how?
Dr. Miriam Huntley: My background is on the technical side. I was doing a PhD, really developing algorithms for analyzing large genomic sequencing data sets. Sequencing data sets are being produced in a lot of different applications. I was really interested in one particular scientific application, but as I was working on that, I was wondering, could we apply some of this to healthcare and what areas are kind of underserved right now in health care? So that was my path. At the same time, one of the folks that I met in academia who ended up becoming one of our co-founders, Doug Kwan, he came in from the medicine side. So, he's an infectious disease physician at MGH. He was really seeing the unmet need where he would have patients who are literally dying because he could not get them the right antibiotic on time because diagnostics were too slow. We came together, and we had this question, could we use genome sequencing as a way to provide a faster diagnostic? And we ended up getting together with a couple other scientific and business folks who ended up becoming the cohesive co-founder group that ended up starting the company. We started looking into, could we really provide a new diagnostic using the new modern technologies to solve a very, very old problem?
Dr. Frank Maddux: How hard is it to move from things that were effectively academic science projects? I mean, very sophisticated science projects to trying to be a commercial entity, which is obviously very different. And I'm just curious, what's that transition feel like for a small company?
Dr. Miriam Huntley: There's a lot of challenges going from a small lab or a small computational program to actually doing something that's commercial. It's robust, it can be used in the real world. It can be used on clinical samples. Challenges like how transfer protocol that's been developed with people and pipettes into an engineered device that works in an automated fashion, and it has the robustness that you need and the precision that you need in order to actually go out. How do you make sure that an algorithm can handle all the edge cases, every edge case that you can think of, you'll get that and more? And so really kind of thinking through all of that. A lot of the challenges have been really how do you go from proof-of-concept which is often what we have in academia to something that, OK, this is not only an idea, but it can actually work in the real world, and it can work for real people who need an answer in real time.
Dr. Frank Maddux: Sample prep is going to have to get a whole lot straightforward for a community hospital lab to be able to do this. Do you think it's doable that you're going to be able to get from these relatively, significant prep steps to something that's straightforward enough for a lab to do? A lab in the community?
Dr. Miriam Huntley: Absolutely. I mean, we're targeting the Holy Grail of sample prep for this problem, which is a sample in-answer out. Automated doesn't require hands on time. We really want to have somebody be able to provide the clinical sample and no touch time and be able to provide the answer to the clinician, HR system 8 hours later. We hope that this gets applied and gets used in many hospital systems.
Dr. Frank Maddux: Has the company had to use the resources of the technology transfer group? I know we did some projects with Eleazar Edelman a few years ago who had been heavily involved in the tech transfer process from a variety of the Harvard related systems. And I'm just wondering whether that was beneficial for you all.
Dr. Miriam Huntley: There was some amount of work that we did with MGH, but I would say one of the biggest legs up for us as a company is when we started, we actually incubated in Harvard's startup ecosystem. So, Harvard just when we were starting the company, Harvard developed this new system called iLab, which was for incubating startups for students and affiliates. And so We started out there, and then just as we were about to leave that area, they developed a new building and program called the Life Lab, which was for companies and biotech to incubate and have wet lab space. It was an incredible opportunity for us as a company to get our feet, and get through proof of concept, get through some validations before having to raise a lot of money, have our own space and really prove it out at a much larger scale.
Dr. Frank Maddux: You're living as a young company, not a pure startup at this point, but a young company through some of the most difficult times in innovation and financing because of the pandemic, the economic environment, and so forth and so on. How hard is it to attract people and get the right skill sets that you need and capabilities? You've probably got some fairly highly specialized comp bio and data science folks, I would imagine.
Dr. Miriam Huntley: We've been incredibly fortunate with the team that we've built today. I think we have some of what I would consider the world's experts and not world experts in a single field, but experts in many different verticals. I think one of the challenges and blessings of the company is we're trying to integrate a lot of different technologies. And what that means is we need experts in each of those verticals. Whether it's expertise in microbial genomics or expertise in cloud-based engineering or expertise in machine learning at scale, we've been able to attract talent who are really world class experts in each of those and enable cross team collaboration between all of these experts so that we're able to have a cohesive technology at the end of the day.
Dr. Frank Maddux: In effect, one of the things you're developing is this opportunity to identify a whole series of biomarkers for specific antibiotic resistant bacteria and recognize them super quickly. What's your sense of how the diagnostics and biomarker space is changing at this point? And where do you think it's going to be ten years from now?
Dr. Miriam Huntley: I'd say the old the old archaic systems are culture. They're the workhorse of microbiology labs, and they can work well in many situations, but they're just very slow. So, that's kind of the old standard. There's a set of technologies that are the new, newer molecular technologies that are out today, and those are PCR-based, and they're they can be very sensitive. But they're very narrow in what they can pick up, so they can pick up a handful of species or a handful of antibiotic resistance genes. But those are the new technologies that are out today. For us, we're targeting the next generation of technologies, the next generation of diagnostics, which we think is sequencing based. Of course, if you could sequence the whole genome of a pathogen, you would. Of course, you just need the enabling technologies that make that able to happen with a signal to noise problems and the back-end compute and algorithms to make it useful. So, we kind of envision the future where PCR is obsolete and really whole genome sequencing is it's what's being used.
Dr. Frank Maddux: What do you think the timeline is for seeing that transition to this next generation of sort of diagnostic testing and infectious diseases?
Dr. Miriam Huntley: You kind of mentioned sequencing technologies and some of the new technological developments. It's been very interesting to watch that industry. As a company, we don't develop our own sequencing technologies. We use off-the-shelf. And so, looking at that industry and seeing how costs have fallen dramatically, there's some graphs we can look at where the cost of sequencing has fallen faster than Moore's Law.. It's one of these areas where it's just becoming cheaper and cheaper. We envision in the next three-to-five years the sweet spot where sequencing becomes cheap enough to be used in infectious disease applications.
Dr. Frank Maddux: How long do you think it'll take where just routinely humans all get sequenced?
Dr. Miriam Huntley: I think not too long. I mean, again, hundred-dollar genome is really on the horizon. And I think it'll be not just how long to humans will get sequenced. We all have that, but how long? How long does any kind of sub-sequencing that you want to do or xome how long does that happen as well?
Dr. Frank Maddux: One of the reasons I wanted to chat with you is I think we're looking at areas of new capabilities in medicine that we think are going to have an impact on kidney disease care. And the thought that our diagnostic and pathologic classification of kidney disease is not nearly as specific as it should be. And in doing this, creating more precise therapies, it comes in a lot of different forms, one of which is how do we treat disorders or identify disorders that our patients get, not just what's the patient got a likelihood to get. This is one of the areas where I think genetics, genomics, and other omics I think are going to become substantially a big part of our lives medically. Are there enough people being trained in the general field to be able to service the amount of stuff that's going to need to be able to be done to support sort of that precision medicine approach?
Dr. Miriam Huntley: Having folks who have a deep understanding of genomic sequencing data and then paired with an understanding of data analysis and machine learning, those are important skill sets. I would say they're relatively rare, but they're growing. I think the confluence of machine learning and biology and biological sequence data is becoming more and more of a hot academic topic. I think the training of kind of in the academic field is increasing, but it's still not saturated. I think there's still a lot of ways to go. One of the things that we found in our interactions with clinicians with hospitals is the need for providing automated interpretable results from genomic sequences. Like I mentioned, no, no clinician wants to really get down into the weeds of what are the nucleotides, right? They want to have an algorithm that works out of the box that they can provide the diagnostic result for them. I think the need for providing these automated algorithms is going to continue to be a really critical one.
Dr. Frank Maddux: Any other areas that you think we should chat about today related to the company or the technology you're developing?
Dr. Miriam Huntley: One area that we didn't touch on is that of hospital-acquired infections. And it's an area that as we think about outbreaks, as we think about the ability of an infection to spread in a group, it's one that is now in the public minds more and more with COVID. With COVID, we're seeing a lot more use of sequencing to better understand how a strain can propagate and whether on a cruise ship or whether it's from a conference. We're actually better able to trace that back. So that's a capability that's being well used specifically in the COVID space, but it's something that hasn't quite yet translated to the bacterial infection space. Healthcare-associated infections, they're a major problem in the United States and elsewhere. But unfortunately, the technologies that are available to hospital infection control teams are pretty limited at the moment. If a clinician expects or in a hospital for infection control person suspects that two infections might be related, they have to kind of look at, well, you know, are they the same bug? Do they have the same clinician, or do they use a shared equipment or are they in the same space at the same time? Our collaborators in this space who might be considered some top of their field in infection control. They've called themselves the gumshoe detectives and infection control because they're really trying to connect the dots. You know, on the flip side of that, if you can use genome sequencing, you can really tell with extremely high levels of accuracy, are these infections related? Is there an outbreak happening or are these infections just happenstance, and they might be a community acquired?
Dr. Frank Maddux: This field of what I'll call forensics, infectious disease forensics is going to be really revolutionized by the kind of work that you all are doing.
Dr. Miriam Huntley: Absolutely. So it's something that we've spent a lot of time investing in and developing solutions around both for the healthcare as it stands right now. So, we work with hospitals to help them sequence suspected outbreaks and determine if things are related or if infections are related or if they're not. And then also, it's the future that we envision. When our diagnostic is being used in hospitals and we're sequencing bacteria on a regular basis, we'll be able to determine, hey, this infection is actually identical to an infection from a different patient that we saw two days ago. And that will help infection control go back and look. OK, they shared the same ventilator and that will kind of help kick off outbreak investigations.
Dr. Frank Maddux: I've been here today with Dr. Miriam Huntley, chief technology officer for Day Zero Diagnostics, talking about a novel use of genomics and genetic interrogation of bacteria. And Miriam, thanks so much for being with us on our Dialogs show.
Dr. Miriam Huntley: Absolutely. Thank you so much for having me.