Slava Akmaev is the Chief Technology Officer at Scipher Medicine. He is responsible for product development in precision medicine and early discovery efforts in drug development using the Network Medicine platform. He has been recognized as a leader in the adoption of the AI/ML technology in healthcare and drug development and is a frequent speaker at some of the most prolific industry events. Slava is the inventor on a number of issued and pending patent applications and has published more than 30 peer-reviewed articles in computational biology, artificial intelligence and molecular biology. Additionally, he authored book chapters and numerous scientific presentations and posters.View the profile
About the talk
Again, thank you so much for attending the session. It's it's getting late. I'm on the east coast and happy to see the few participants here. I'm interested in learning more about Cipher going to share my slides real quick, see if it's going to work. and the switching to the presentation view here You should be fine. Yeah, and I think we can begin. Soyeon, my name is Lala Acme. If I'm the chief technology officer at side of her medicine, I've been in the industry for more than two decades and the last 10 years specifically focused on artificial intelligence and machine
learning in life sciences. And also Healthcare using molecular data and also data that is more digital in nature such as patient outcomes data claims data are the types of data that this role into real patient care in in healthcare system today and how we use Ai and machine learning at Cipher to develop new products weather Therapeutics or Diagnostics and additional. How do we utilize experimental data on top of it and combine some of the machine learning
technology it with the experimental biological beta Couple words for outside for a small company located in Waltham, Massachusetts where diagnostic company and we have a CLIA laboratory in in Durham, North Carolina. We just recently launched that test and rheumatoid arthritis called Prisma Ray and I believe it's a revolutionary test. It's a truly Precision medicine test that help answer the question of response and the normal response for some of the very expensive Therapeutics and rheumatoid arthritis.
Moving on a couple of words about our Founders. We've been found a few years ago by Joseph loscalzo who is chief of medicine at Brigham and Women's Hospital here in the Boston area and my professor lost libido by zip. He's a physicist and he's been in computational biology for over two decades Hughes director of complex Network research at Northeastern University. Our backers the company is I said been around for a few years and we have multiple rounds of funding. We have some of the well-known backers in
healthcare industry Ventures khosla Ventures northpond Ventures. We back by United Health Group as one of the largest health insurance in in the United States and in a few others So what is it that we're trying to do? We're primarily focused on autoimmune disorders and I'll talk more about rheumatoid arthritis today. The idea is that this space of autoimmunity is a little bit of Uncharted territories when there's quite a bit of development that happened in the last 15 years were now we understand that
patients with tumors to respond to therapy it primarily depending on the composition of their humor and humour driving vacations. And that's how we begin to select specific targeted therapy for oncology patients. In other diseases. It's not so Advanced and specifically in autoimmunity. They're practically no Precision medicine test no predictive testing capabilities upload identify and match patients with a was therapy. So what we're trying to do at Cipher at least to find these letter signatures, Would really pinpoint specific therapy
the right therapy for the right patient. Movie home again just a couple of words are in very general about how we approach dysregulation in autoimmunity. And how does it compare? It's really nice to have a parallel and compared to oncology. Right? We understand now that a lot of the humor of Genesis is driven by mutations into Mercer's somatic mutations and a lot of the development of diagnostic testing Sora walls around how can we find the somatic mutations weather in tumor
biopsies or potentially liquid biopsies the blood testing and then identify the right therapist diagnose Jumer's improve patient care through the type of approach, you know by this time. It's I want to say straightforward but it's well understood the path for precision medicine in oncology is well accepted in the community. Spell accepted in the ankle centers Medical Center in research. It's a little bit different and it may be in the way more complicated because we
still have an autoimmune disorders aerobics else ride the cell immune cells that attack its own horse right away is kind of similar to tumor would he have uncontrolled growth of cells come from joint damage or other types of deficiencies in in the human body, right? So how do we identify these cells because there's nothing wrong with their DNA. The DNA of the cells is exactly the same as the DNA or of every other cell in in the Bible. I'm sensitive looking at DNA. We have to look at other types
of measurements and signal In the cellar email you and specifically will look at Dynamic measurements such as opposed to DNA molecular data. Just driving in expression. Can the gene expression cell looking at RNA expression measuring it in this earth roses that Dynamic state it can be protein expression and Pregnant. Which is the next stage. So to figure out what is happening in autoimmune diseases. We need to look Beyond DNA and we need to understand a much different set of variables and features
and ability to measure them to make an impact in inpatient care. I'm so what we've done in cooperation with our academic Founders and partners is that we develop AI data-driven platform that uses experimental data obtained through thousand sin and sometimes millions of experimental points where we put it all together into this network of the cold human interact on in this network represents physical protein protein interactions with been verified through biological experimentation and what we're looking at right
now and in our development we can we can imitate 18,000 proteins in the human cell through this interactome and then have more than 300,000 connections representing these physical interactions between these 18,000 protein. This work has been submitted this technology has been in the works for more than 15 years fitzwell published. We have papers in in nature science and other high-profile journals and They deserved the academic drivers and some of the scientific drivers of this technology are helping us
to commercialize in the lead for first completely normal inside in our immune disorders. So what what is the unmet need and what is the question by starting research has been uncovered that about 225 million patients in the world. Don't respond to don't respond to their prescription medication. What does it mean? It means that people are prescribed drugs can range anywhere from blood thinners all the way to very expensive targeted treatment scorpion oncology or autoimmune disorders and
patients take these medications, they subscribe and they can send for potential side effects and they practically have no benefit from the street. So how do we identify the individuals that are likely to respond? That is the biggest challenge right? And this is what we're focused on that site for men. A bit of a technical aspect of it everything that we do is represented through networks. What I have here is just a small part of a sub Network. What's the it's imagine is a subnet some of the biological signalling or mechanism of action of sound of
compounds in rheumatoid arthritis. So then through some of the gene expression data in experimental data that we can obtain in additional experiments and and collaborations with Academia or medical center. We can identify biomarkers, but may represent patient subgroups that are likely to respond to certain pregnant. Right and these two groups are represented by the old host of the Earth around the middle of a patient's upper be patient. Proceed, right? All the subgroups in there biomarkers can be
overweight on top of the network and then what happens that using Ai and machine learning. We can analyze the proximity of non therapeutic targets weather is for example, anti-tnf treatment for rheumatoid arthritis and the by understanding the topology in the relationship between the network of the propagation of the signal from the target to the patient's subgroup. We can make decisions by using Ai and actually a is making decisions for us whether the patient is a good candidate for that specific treatment. In addition to
the store of high-level approaches where we identified disease module swear that if I left work published in the papers that looked at finding therapeutic targets finding compounds and refurbishing opportunities for compost already on the market identify drug combinations that may be more important than a specific therapeutic that were looking at all. So as I said looking at patient individual data in the context of this is Marshall identify patient classification certification opportunities looking at Precision Therapeutics, and I'm the last thing that I wanted to mention here is
probably the most complex challenge in in biology and Life Sciences is to understand biological network analyst and biological Pathways in Context of perturbations in the context of the Seal of the simulation who published number of papers on perturbation specifically in the brain and other organs that would be able to help the researchers in the medical community to Alyssa date in silico. What interventions are more appropriate for specific phenotypes what interventions can be favorable for people with certain
malfunctions and dysregulation molecular pathways. Do all these papers are available in the public domain always late and many of these algorithms are available online with an addition to that aside for medicine that develop their own proprietary algorithms or even be these types of approaches. Now, I want to spend a few minutes on our first product that will launched early this year. It's called prismarine it is as I said first time ever that we have a predictive test a rheumatoid arthritis and the this test read a person's
power of personalized medicine in a in a somewhat prevalent disease rheumatoid arthritis in the United States and allows us to divert patients from an S surgery save a cost of treating patients with the therapeutic sub don't work and also improve patient outcomes. so now in summary this prisoner a test that is developed by using artificial intelligence and machine learning and specifically the test itself using a random Forest algorithm combining multiple
pictures into one score allows us to change some of the phone formulary that is present in your ass and for many of your Cupertino. Familiar with irate patients that are about to go on a targeted therapy people care prescribe the class of medication for cold anti-tnf. These are medications that inhibit tnf NFL function and they allow certain patients to reduce the level certain cytokines in in their circulation the surf Amelia team some of the effects of the Overture to be in system not
everybody response and we'll talk a little bit about more in full of slides, but the formulary that are presented by a lot of the insurance companies. Physicians mandate that almost everyone should take an anti gnfs their first line and go through cycling up anti-tnf until they go through to three cycles of different anti-tnf in Absolut liquor. The decision is absolutely convinced that they cannot respond to the medication. So by using Prisma Ray and diverting people from on this surgery is next weekend in increase the response rate by up to 40% and you will see that we can
reduce oral health care costs and avoid expensive in the effective therapy. So how did we start on this project? It was a really interesting beginning because they are still going to diagnostic companies. We started this journey by talking to some of the biggest insurance companies in the United States and we asked the question. What is the most critical need for the Paris right now in dealing with some of the runaway cost and in health care and what is the Korean word is used they look at and what we heard from almost
unanimously is that rheumatoid arthritis in anti-tnf medications with some of the most impactful aspect of their health care coverage to their experience right now. So we partnered with some of them and your partners with Rheumatology communities to work and develop this product called prisoner a using aai and experimental data that we have throttle platform. No. What is the actual need in rheumatoid arthritis? I talked about no response turns out that this anti-tnf treatments are extremely expensive. What you see is
33 billion dollars spent in the United States on anti-tnf and it's probably some of the world's largest selling drugs. They're very expensive course of treatment can range from somewhere in about 30 to 40 thousand dollars for just a few months. What's interesting here is that two out of three patients approximately do not adequately respond to that medication. So they may go on a medication standard for sale 6-9 months and have no. No positive effect from the increase in seeing him spend as you can see in the slide on the left is increased from
2012 to 2019 by almost 200% with practically zero changing the response rate, right? So we're increasing the use of this medication. We still don't understand which people are patients are likely to respond who should get that medication is the first line and also it's worth mentioning that in rheumatoid arthritis options as first-line treatment that are alternative treatments to the tnf Inhibitors. There are many other really very similar to the case of compounds that can be used as Target the treatments for for these patients.
So the priests marry when she used to test recently, it's a 23 biomarker panel. We use 10 Single nucleotide polymorphisms. And we we have a transcript in the panel and if I clinical features as I said, it's combined through machine learning algorithm call Grandma forced into your lawn's scoring function will we have extremely optimistic and end the enthusiastic adoption of some of the caramel sundae in the rheumatoid arthritis? They see Prisma Ray as a for a Precision medicine into all the immunity in the right space in particular validated this test in hundred
seventy-five patient population of Bullet by Lynch play for patients that have not experienced targeted treatment in rheumatoid arthritis from the largest are a registered in in the space the coronavirus play. Maybe not that such a great name these days but Corona Registries really famous and greatest. They were able to collect information on patient outcomes from tens of thousands of patients. And in addition to that they were able to run a study it to collect biological material Baseline blood samples or blood samples during
treatment from thousands of RA patients with partnered with the registry to develop and validate. Our technology is you can see it on the left. The predictive power of dust is rather High. We have pasta predictive value of my uncle almost 90% specificity of almost 87% and the odds ratio of predicting inadequate response. In order of being an adequate responders responder in the predicted United response category compared to being an adequate responder in the predicted on Durant category is 6.6, which is extremely
significant. Now when will look at the clinical utility of prism or a beef with stratified patients before therapy using Prisma rate test, then we can increase response rate for these patients by almost 40% right going from about 30% response rate that we observed in our Corona collaboration to 43% If we actually had to use Prisma registrada for these patients to anti-tnf therapy were alternative Christmas. So this this new paradigm of precision medicine actually allows us to complete this shift the structure of in proportions of responders and don't respond. That's what you see on the left.
And in the pie chart is about 30% of people in the orange are going to respond to anti-tnf any up 70% of the population that are not responding to the right if we implement prismarine this what we're doing right now very actively with our academic and Commercial partners. Then we're changing this entire distribution right now. We have about 26% of tnf responders. We have responded to Alternative treatments which about 16% all in all combining to 42% response rate and then steal that we have a supply of the
population that is not responding which requires additional work on other treatment mechanisms in identifying very similar predictive Diagnostics what we can pin. Went exactly which therapeutic they should go on. And then looking at the health economic impacts what you see in the headline is that by using his mother a wheel reviews of the net p.m. P.m. Was just per member per month cost by 11 sets. Right and it may not sound much. But if you have a 11 million patient plan, and this is for every member with a person
having rheumatoid arthritis or not. This is a significant number that accumulates over the years and if you look on the right side overall for for a plan with 1 million members were saving about 7.3 million dollars and cost with sending $6,000 per test a patient for a year and we're saving 5.9 million dollars in an effective spend which is extremely significant. So just to summarize. this ability to predict treatments and ability to actually connect patients with the right treatment is is it would enable us to transition from a rebate driven perspective on the health
insurance side, which means that certain Therapeutics are mandated as first-line and insurance companies are encouraged to to use those Therapeutics by the black manufacturers by providing soaking rebates based on the volume of Therapeutics used, right if there is no Precision Diagnostics, it's extremely difficult to fight the system. But if we have Precision Diagnostics predictive test and we can show the validation studies that these tests are working and we can predict what I listen to the
significant with eyelet statistical significance. correctly for the right treatment, then we can transition from the rebate during perspective to Precision medicine formulary which how we believe Healthcare should be Going to summarize here with our commitment and be happy. We have a few minutes. I think to answer questions. Sorry, I can barely hear you. Are we had some more time? So I have a question in chat or is asking is it possible to talk a little bit about
particular technological stack behind the platform? Yes. We use everything that we do with our data sets is an Amazon AWS Cloud. We use a lot of data analytics tools and some of the tools that are open source for driving machine learning and AI specifically neural networks and random Forest algorithm. And in addition to that we use some of the open-source capabilities to manage our extremely voluminous data.
Buy this talk
Buy this video
With ConferenceCast.tv, you get access to our library of the world's best conference talks.