Big Compute - HPC and Genomics Revolutionize Medicine

Episode Date: May 16, 2019

Gabriel Broner hosts Mark Borodkin, COO of Bionano Genomics, to discuss how genomics and HPC enable doctors and researchers to diagnose complex diseases and prescribe unique personalized treatments based on individual variations of the DNA.

Hello, I am Gabriel Bronner, and this is the Big Compute Podcast. Today's topic is improving diagnosis of complex diseases. There is new understanding that correlates variations in the DNA with the onset of diseases. Doctors may be able to run tests, catch diseases early, start treating patients, or develop new treatments. We can imagine a day when we go to a doctor's office. They take a DNA sample, run tests, tell us what is wrong, and what is the best personalized treatment for us. But what are the challenges to make this a reality? To discuss this new approach to diagnose diseases, our guest today is Mark Borodkin. Mark is Chief Operating

Officer at Bionano Genomics. Bionano uses HPC in the cloud to improve the diagnosis of complex diseases. Welcome, Mark, to the Big Compute Podcast. Thank you, Gabriel. Happy to be here. Yeah, it's a pleasure to have you, and we're all interested in this new topic that is evolving. Mark, BioNano focuses on complex diseases, so if we wanted to start from the beginning, can you share with us what complex diseases are? Yeah, sure. So typically I lump them into two distinct types of diseases or disorders.

There are what we call the rare disorders, which typically occur in very small subsets of the population. Less than 200,000 people may have a particular disorder however there are over 7,000 of these types of individual disorders that we know about and in just the US alone there are over 25 million people living with a rare disease. So they are not very rare when you look at them in total. And then the other area that we focus on as complex are cancers, which unfortunately are not rare, but where each patient has a unique cancer signature and that cancer signature can

impact how that cancer progresses in that patient and what treatments work on that on that patient okay so many of the disease I actually went to a nature magazine to look a little bit about complex diseases. It seems like things we know like asthma, Parkinson's disease, Alzheimer's disease, many of those could be considered complex diseases. Is that also correct? Yes, there are diseases also that we classify as undiagnosed diseases so for example something like like autism where it's a it's a very complex disease we can clinically tell whether a child has autism however we don't have a definitive kind of molecular or genetic signature for the different

classifications of of the autism so it is not it is not rare is not a cancer but it is in all intents and purposes an undiagnosed disease. That's great. Good to understand. So for the diseases you focus on, can you tell us how the DNA changes when one of these diseases happens? Yeah, sure. So I guess let's start at the beginning. So our DNA that's captured in our chromosomes that encodes for who we are and what happens to us at a genetic level.

So that genome is altered constantly. So there is constant changes that happens in our genome. But majority of those changes cause no harm they could be certain let's let's call them spelling mistakes and in words that really don't don't code for a particular protein that does that does anything of any significance however if those alterations or variations as we call them if they do for example delete a gene then that function that that gene served to to provide will be missing. And sometimes our bodies are able to compensate for that missing function,

but often we cannot compensate for it. There is a type of muscular dystrophy, for example, that impacts boys, which is fatal, which causes a deletion of what's called a dystrophin gene. So that's a loss of function of the ability for muscles to thrive. There are other alterations or variations where genes that are very far apart maybe on different chromosomes can be fused together to create a new never-before-seen protein in the body and that protein can cause harm such as there is a type of leukemia called CML

and such a fusion drives the the the onset and progression of that of that leukemia and there are many many other types of structural alterations which which in totality we when they're large, we call structural variations. So there's a lot happening in us. So a lot of changes happen all the time. Some of them are not a big deal, but some of them can cause some of these diseases that could be fatal. That's right. Yeah. So can you tell us how we can go about detecting these variations in DNA to improve diagnosis then? Yeah, certainly. So the detection of

large variations, we've been doing for the last 50 or so years with classical cytogenetics techniques. Cytogenetics is the field of medical science that these fall under. Some of the earliest ones that are still very much used today for lots of different types of diseases are called karyotyping where you're looking at and of chromosomes under microscope microscope similarly technique called southern blot and more modern techniques that were introduced just in the last couple decades fluorescence in situ hybridization and array CGH. Most recently, DNA sequencing has been introduced,

and its power is really to look at finding small variations and less complex variations. And most recently in the last few years bionanogenomics, my company has introduced optical mapping with the attempt to replace these classical cytogenetic techniques for these large variations and doing so in an industrial scale way. So in the things you do today, you call them optical mapping. Can you tell us how those are different or in which way they help you with the diagnosis compared with other techniques? Yeah, sure.

So what we do is we take very long pieces of DNA. We extract those from the cells of the human body, whether it's from blood or tissue. And we were able to prepare these DNA molecules with a labeling chemistry that we have, a fluorescent labeling chemistry, and we're able to then interrogate that DNA and look for for patterns that are different than let's say a healthy individual or or a healthy part of the patient and and so we're able to find various structural variations against kind of

those those known references so you have it like a table of the types of variations that correlate with diseases and you try to find if any of these variations is happening on this particular patient is that how it works so so yes we could we could both look for variations that have already been discovered but we oftentimes people choose to use optical mapping to discover new variations that have never been seen before to determine new markers for developing drugs or other therapeutic interventions. I see. So one of the promises of the genomics that we're seeing today would be that we go to the doctor, they understand what changes are happening, and they could also be able to prescribe a very particular treatment, correct? Depending on my DNA, there may be types of drugs that work better or have fewer side effects is that fair yes yes you know just you know look at cancer as an example you know we we say you know a person has a particular type of cancer by you know

calling calling out where it occurs so you So we may say a person has a lung cancer, and there are certain therapies that are prescribed for a lung cancer patient, and yet there are folks that respond well to that therapy, but then there are folks that respond very poorly to that therapy, and yet they have the same cancer as we defined it from the tissue of origin. Why is that? Cancer researchers want to understand that. And what we've been realizing in the recent past is that each of these cancers, even though they may be in the lung, has a different signature, a different set of variations that causes that cancer to progress differently, to respond to therapies differently. And so it is those signatures that we're trying to derive, to discover. And really kind of when you think about the goal of kind of this new wave of therapies is to be able to prescribe a therapy personalized to you based on your uniqueness or your cancer genome's uniqueness. That's very interesting. So you can detect the cancer, but you can prescribe a more effective

treatment by understanding which type of lung cancer this is, for example. That's right. Yeah, that's very good. So you have developed solutions for disease diagnostics at BioNano. Can you tell us a little bit about them? Yes, so we currently have or are used in a number of clinical studies. So we are not yet deployed in the clinic as a test, but we are applied on patients worldwide in an alternative research workflow. And those clinical studies involve various leukemia. Leukemias, they also involve various muscular dystrophies. And in some of those studies, researchers are interested in just replacing current standard of care that uses classical cytogenetics with something

that is cheaper, faster, and provides a higher resolution of the disease, of disease indication that they're looking for. And in some of these studies, they're looking at identifying novel biomarkers. So, for example, you know, we talked about rare disease diagnostics. Well, you know, when a patient presents with something that looks like a rare disease, kind of from the clinical presentation, the success rate of actually being able to diagnose that with any test ranges from 30% to 50%, depending on what indication that is. So for the others, for the other patients, there is no biomarker that's discovered in terms of what's causing that genetic disease so folks are trying to find those biomarkers because these biomarkers are kind of the predecessors of the predecessor steps for developing a novel therapy right if you don't know what what's causing that

disease you can't develop a therapy for it or a drug for it. That sounds very good. Are there any early results that you're seeing or any promising results you've seen in these trials at this point? So there's been a number of publications that our customers have made. Most recently, there were a set of publications around diagnosing a very difficult to diagnose muscular dystrophy called FSHD. I'll spare you with the medical term for that.

But it's a muscular dystrophy that affects individuals in various ways. Some of them could be very debilitating and some could be milder. And there are a number of markers that we can detect that allows for the diagnosis of that disease in a more refined way, in a faster way. And we're hoping that will be adopted in the clinic soon.

That's very good. So you gave us a good view of the kind of things that you're able to accomplish with these new products. I wonder if you want to say anything more about the challenges that, in general, modern genomic tools are hoping to solve? Yeah, I think we're really at the beginnings of using genomics, genomic tools to understand these complex diseases. We're still very early. One of the things that we and many researchers are discovering is that often we are overwhelmed by the information that we see. We detect both with BioNano but with other technologies. technologies, we as a community detect very many variations in the genome that look unique to patients, but which of those are actually clinically relevant to that disease. So that is

looking for needles and haystacks. And often clinicians get overwhelmed by the number of signals that are discovered for their patient. And what they want to know is which of those are actually important to that particular disease. And so there's a number of these disease association studies to try to associate different diseases with different variants. And so that is a very, it's quite a bit of work to do that. And a number of very large studies around the world are occurring today to be able to do that. A number of very large studies around the world are occurring today to be able to do that. Oh, that's good. You're telling us we're at the beginning of something, really. The more we advance, the more we'll be able to look at variations in the DNA and be able to understand the specific

diseases, the specific treatments, et cetera. Can we imagine a future where we get to the doctor's office and the doctor immediately runs a DNA test on us and tell us what's wrong with us, what's the treatment? What are the challenges to get there? I think, so that future will come. That future will certainly come. The challenges that we're foreseeing in order to realize that are,

some of them are technology challenges, you know, the tools that we create need to get ever faster and the results need to be generated with less and less cost. But those are beyond those technology challenges, there are actually far bigger challenges that we as a community need to work through. be able to convey the information that's found in your genome in the proper context, right? Because sometimes we can say fairly definitively that, okay, this variation means that you have this type of a disease. But more often, it's usually a probability. So this variation creates a certain probability of you getting this disease or you having a particular issue.

And trying to convey those types of probabilities in a clinically relevant way is done by genetic counselors and frankly we're going to have a shortage of genetic counselors and you know there are a lot of companies today that are looking at AI as being a way to to help with genetic counselors to be able to to you know to to support a greater population of patients so in this case say I would be used when you get a genetic result AI will give you not only a probability but will suggest the course of action to you? Something like that?

That's right, that's right. So that would, that's a, that's one of the key things that people are looking for. AI can also help with with things like understanding the research literature to derive or help derive kind of new ways of looking at that particular disease, figuring out what therapeutics would be most suitable for that particular signature. Right. So I know, for example, it's been used, AI has been used in radiology, right, to understand if there's a disease that can be found on an image. Similarly, we could use AI to start finding correlations between gene modifications and diseases. Is that a thought? Yeah, absolutely.

Yeah. Absolutely. A lot of work going into that. Yes. No, that's good. And the part that you talk about counseling, I mean, there's been a lot of advertised cases, publicized cases of people that get a DNA test and decide a course of action, assuming they're going to get breast cancer or something like that. Is that where counseling becomes a key important thing

after getting a DNA test? Yeah, absolutely. You know, there is, I guess there's a lot of empowerment with knowing, you know, knowing more about your genetic predispositions to a particular disease. However, with that comes some uncertainty and so genetic counselors help to navigate those waters with patients. That is definitely interesting. We'll look at all the possibilities of the future of understanding our DNA will bring.

Mark, let's switch gears a little. You chose to use HPC in the cloud for DNA analysis. Can you talk to us about that choice and how you are using HPC in the cloud today to make this a reality? Yeah, certainly. I think we came to a realization a couple of years ago that we are far outpacing Moore's law. So we're outpacing traditional compute, both by our ability to generate patient data or collect patient data, the genomics of patients more quickly, but also in our ability to look at more and more complex events in those DNA molecules. So we're able to process more patients, and we need more compute to look at more complex events in the genome. And so we looked at HPC in the cloud as the best way for us to be able to scale

with what we're doing today and what we envision to be doing in the future so that that allows us to scale more quickly you know and answer the questions for far more quickly than we could with with traditional compute so effectively you take a DNA sample you do some processing and then you submit that to your HPC in the cloud platform is that how it works yeah that's that's right that's right and so we we've chosen to have kind of been it's called an integrated HPC in the cloud solution with rescale and so that allows us to as soon as we get the data off of our instruments to be able to push that into into the cloud and in the cloud resides

our bioinformatics pipeline which then crunches through that data and allows us to get results much faster for a large number or large cohorts of patients at the same time. Very good. So it's very interesting, Mark. I mean, clearly you're pushing the boundaries. You're looking at a field that is nascent, like genomics, and understanding complex diseases and the right treatments. Are there any learnings from the process you're going through, both in pushing the boundaries in your field and in pushing the boundaries of computing of what you're doing?

Anything you'd like to share for people listening that could be in your area or other areas that may get inspired by what other people are doing? Yeah, I think one of the things that we realized pretty early on as we were researching which way to go with the cloud was that we really wanted to provide a solution to our researchers and clinicians that is fairly easy to use. So one option that we had in the beginning was just to, let's say, do it ourselves, right? So put our pipeline in a container, package it on AWS or one of the other cloud providers, and write some application notes and have our users go at it and use that with their own accounts.

And what we realize very quickly is that the vast majority of our customers will be very turned off by that because their job is not to understand the jargon of the cloud. Their job is just to get data so that they can make insightful findings of their patients. So we decided to really abstract the HPC and the cloud as much as possible for them through our software and through the rescale platform. And so that our customers, all they need to know is the security and the performance guarantees that we make. And the computing gets down in the cloud and all their data gets returned. And they don't have to worry about the new ways that Amazon or Google are going to be

doing the new jargon that they're going to introduce tomorrow. Yeah, it makes sense. So your customers really would do genomics work and they don't have to worry about computing and that's a plus for them. Yeah, that's right. Yeah, that's very good. Listen, Mark, it's been very interesting to listen to you. This idea that we go to the doctor, they take a DNA sample, they understand how our DNA has changed. From there, detect exactly the disease we have and also what's the best treatment for our changes seems to change medicine, right? So the future of medicine is not going to be what it used to be,

that as you said at some point, lung cancer has a particular treatment, but it may work for some and not for others. We've got to be much more precise with this idea of more personalized medicine. So we look forward to the future, and I think we'll get to see it as doctors start using more and more of the tools you and others are working on. Before we close, is there anything you'd like to add for our audience? Yeah, I just would like to say that I think our transition to the cloud, providing this cloud offering to the customers, was certainly enabled by the unique capabilities that rescale has uh you guys are you know made our lives uh easier in this transition uh and um you know we are definitely

looking forward and and using using this offering to uh to its full potential well i appreciate your kind words i'll i'll relay them to the team that made that happen. So thank you very much. So Mark, it's been a pleasure talking to you. And so basically to close, I'd like to thank our guest, Mark Brodkin, Chief Operating Officer at Bionanogenomics, to help us see a better future thanks to advances in the understanding of our DNA and in the tools that significantly change our ability to diagnose complex diseases. Till next time, I am Gabriel Bronner and this was the Big Compute Podcast. Thank you.