Jon Peterson: That video gave a sense of where we would like to get to in the future. But I’m going to talk a little bit about where we are now and what will happen next.

So our ultimate realization is a digital representation of human physiology, of you, of me, but it turns out that humans are rather complicated. We’ve got layered structures and interactions. So getting a fully realized bio digital twin will take some doing.

So therefore our journey is going to start in the cardiovascular system. Now why the cardiovascular system, well, first it connects every other organ system in the body. But another reason is that diseases of the heart and the circulatory system are the leading cause of death in both the world and in the United States. In the US, 20% of deaths are caused by ischemic heart disease. So we feel that we can have an impact in this area.

Now, even the cardiovascular system is complex with a network of pumps and tubes and valves interconnecting everywhere and each of those parts of the system is responsive to both local and systemic inputs. And so our beginning step is going to be to simplify that a little bit. And here is our simplification to start with, this is a mechanistic model, meaning that every component of this has a biological correlate to it. So this is an electrical schematic, but it is an analog for a fluid based system.

I’m going to walk through this a little bit, but you could see, those of you who have any engineering training, there are capacitors and diodes and resistors and that’s it. So we could start here with the left atrium and we’re representing that as a capacitor because the capacitor can hold charge, or in our view, fluid. And it’s a variable capacitor, and that actually turns it into a pump. And I’ll show you what that looks like in a minute. Well, blood will flow from that through some resistance through the mitral valve. So we have that simulated as a diode and a resistor. And we can allow backwards regurgitant flow by having a diode positioned in the other direction. So this would represent mitral regurgitation in the real world.

Well, that’s the left atrium. The left ventricle has got the same structure as the left atrium. We’ve got the valve and we’ve got a time varying capacitance. I may refer to that as an elastance every now and then, because that’s how the researchers refer to it. But the parameters that are associated with those resistors and capacitors of course, are going to be different from the ones that are in the left atrium.

Well, the blood leaves the heart, goes through the aortic root, and makes it into the systemic vasculature. Well, our arteries in this first part contain blood and they pose some resistance to flow. So that’s how we model it; we’ve got a capacitor to contain blood and we’ve got a resistor to oppose flow. The blood then goes into the capillary system, which is the system across all organs that allows for exchange of oxygen and nutrients and takes away waste.

And then finally we make it into the venous system and you’ll notice I’ve changed colors on the schematic here. And this is just a qualitative representation that we’re going from oxygenated blood to deoxygenated blood because through the capillary system, the organs have taken out the oxygen that they need.

So that deoxygenated blood then makes it into the right side of the heart. And again, the same structure as the left side, just different parameters. That’ll then make it through the pulmonary system, the lungs, where the blood gets reoxygenated and the circulation starts again.

Well, this may look like a simple system, but there’s more than 100 years of research going into this. We’re standing on the shoulders of giants in this. And to honor one of those people, Otto Frank, who was researching and publishing in the field a hundred years ago, we’ve named our model Otto. So meet Otto.

So we’ve implemented Otto in Simulink with a MATLAB front end, for those of you who are interested in that. And what I thought I would do here is just run through sort of a geeky technical simulation of this allowing for our basic research roots here.

What we do to start the system here is, I’ve got my little circles there to show that we’re going to charge the system up. We’re going to put fluid into the system, in the capillary bed. And then when I turn on the simulation, that fluid will route its way around the body. Now, of course a real body doesn’t start up that way, right? But it’s interesting to watch the waveform progressions as this goes and so we’ll see what happens.

So when I started, you can see that heart rate is at 60. We set it that way. But you can see the blood pressure systolic over diastolic pressure is increasing as is cardiac output, which is the lower number, which is the amount of fluid that’s flowing through the heart per time.

Up here, we’re showing pressure waveforms. The red is the left side of the heart. The blue is the right side of the heart, and you can see aortic pressure in yellow as well. We’ve got all four of the valves represented flow going through the valves there. And finally on the left, we have a phase plot, a pressure volume plot. And this actually allows us to compute the external work, the amount of work that the heart has to perform in order to push blood. And that allows us to compute the metabolic needs of the heart. And while we’re not there yet, this will allow us to explore coronary problems like myocardial infarctions and that sort of thing.

Well, I wanted to zoom in on one cardiac cycle here. This is the last one in that simulation, just to walk through what’s happening, even in this relatively simple model.

So at the beginning of the cycle, we’re just going to focus on the left side here. So the red is the left atrium and left ventricle, and then the yellow will be the aortic pressure. So the left atrium contracts, and since now, the left atrium is at a higher pressure point than the left ventricle, blood flows through the mitral valve, into the left ventricle and on the pressure volume curve on the lower part there that actually finishes filling the heart. So the heart has now got as much blood in it as it’s going to get before it starts to contract. And here it contracts, the pressure rises pretty dramatically, but there’s no change in flow. And there’s no change in volume because all of the valves are closed. So we call this isovolumic or isovolumetric contraction.

At some point, the left ventricle has developed enough pressure that that pressure exceeds what’s in the aorta. And so now the aortic valve opens and you’ve got blood flowing through the ventricle, into the systemic vasculature. And you can see in the pressure volume plot that volume now decreases, which makes sense, where blood is flowing out of the heart. So finally, the left ventricle, when it starts to relax will have lower pressure than the aorta, that aortic valve will close, and the heart will relax with no flow and no change in volume. So again, an isovolumic or isovolumetric relaxation. And finally that left ventricle will, the pressure will go below the left atrium. That allows flow again from the atria to the ventricle, and see you have a passive flowing that actually represents a fair amount of the volume that enters into the heart. And now it’s prepped again for another cycle.

Well, that was a generic model with a generic set of parameters to make it work and give an illustration. But what we want is we want to sculpt this model to define these parameters in such a way that they represent you or me in the context of what’s happening in your cardiovascular system. And we think we have a pretty interesting way to go about doing that. But in order to do that, let me just sort of walk through the inputs and the outputs of the model in a very general sense.

So first we have those parameters that define the behavior of the model, so values for those resistors and capacitors. And we are going to call that a hypothesis space. Each of those parameters is going to have a range of values that make sense physiologically. And we’re actually collaborating, Gomi-san had mentioned some of the collaborations around the planet, and we’re collaborating with the National Cerebral and Cardiovascular Institute in Japan to work through some of these parameter spaces, to define the ranges and to look at interactions between them all. Well, that cube represents what you might see with three parameters. There might be some surface, some volume, in there that indicates what valid parameters might be. Well, we’ve got more than three. So imagine a 20-dimensional hypercube that would contain all of the parameters that we are going to need to search through if we want to find the right one for you.

So now what we have is we have a scenario, so something is acting on you. And it could be as long term as age. That will be one that we’d like to get to at some point. But it could be something much shorter term, which is really what we’re looking at in the near future, which is a medical event, such as a myocardial infarction, a heart attack, or the effect of a drug on you. And so, and of course you can have several of those occurring at once, and that is our scenario space. So finally, we designed the model in such a way that we get outputs, of course, and those outputs are set up so that they are correlates to what we can measure in the real world. Of course, we can also get outputs out that we can’t measure and that can allow us to predict some pretty exciting things about what’s going on.

But for the purposes of this, we’re looking to generate things that we can then compare with what’s in the real world. So we run our simulation across, we’ll call it several million different combinations of values within the hypothesis and the scenario space. And that finally gets us to our measurement space, which is again, a higher dimensional space that we can work through here.

So how do we take all of this and make sense of it? Well, in the real world, as we’ll say, right, we don’t know what the parameters are that specify you in this model. What we do know is the measurement outputs that we get. If you’re in a hospital, you’re getting pressure and waveforms, and that’s what I showed in the model, right? You can measure cardiac output, we can measure heart rate, that sort of thing.

And you are in the hospital for a reason. There’s a scenario that’s associated with that. But those measurements aren’t completely accurate. And that scenario is not necessarily all that well known. So there’s a fair amount of uncertainty in the system.

Well, we’re partnering with a company here in Silicon Valley called Embody Biosciences and their insight is that there are probabilistic connections between all of these hyperspaces. So we’ll walk through a scenario and I’ll give you a sense of what this might look like. So we have a patient in the hospital they’re in there for some medical event. That’s our scenario. We get some outputs. So we get some measurements on this patient, maybe it’s pressure or cardiac output, that sort of thing. Those are our measurements that we have to work with.

Well now what we can do is we can apply a statistical technique known as Bayesian Inference to arrive at some little portion of that hypothesis space that can make sense of all of the data that we have there. And you’ll notice that I have drawn that tiny little part there, not as one little dot, but it’s a little surface because there’s a level of uncertainty in the measurements. But there’s also a level of uncertainty in the parameters that appropriately fit you.

Introducing then that person’s bio digital twin. So now we can take this bio digital twin and we can act on it. So we can play some what-if scenarios. We have this patient in the hospital and the clinical team would like to determine the best treatment regimen for this patient.

So let’s say that they’re considering some sort of an IV treatment for them. Well, with this system, we can apply Bayesian again, Bayesian analytics to start with that patient’s hypothesis space and that measurement, or that drug that we’re putting in, and we can forecast the effect of that drug on the patient. And you’ll notice I’m sort of drawing a little distribution across the side there, because again, this is a probabilistic thing. You got your mean value of what might happen, but there’s a spread on either side. And what that allows you to do, actually, in some sense, this sounds bad, right? You want to have certainty, but on the other hand, there’s nothing certain in life. And this allows you to examine a range of options and make sure that you’ve covered all your bases. So there’s no outlier out here that you hadn’t considered. That’s all built into the system here.

So you can look at this one drug, perhaps you’ve changed the dosage on that drug. You run it again, see what’s going on. Maybe you combine drugs. We after all have a high dimensional scenario space here, so we can do different things to it. And you can ultimately come up with an appropriate treatment for this patient. And by ultimately you mean in five seconds, right? Because we have all of this, we just input the information and out will come the appropriate result. At least that’s the plan eventually.

But there’s another aspect of this that’s quite interesting. And that is, let’s say for the sake of argument that for this given patient, in this given scenario, there is not a pharmacological treatment that is appropriate for this patient. Well, the system can tell you that. And at that point, the clinician can then pivot to an alternative therapy, a device therapy, or something like that, without ever endangering the patient, by going through a time-consuming step of adding drugs and realizing that the result is not what they want to have.

So here’s where we are now. I got to have my little cartoons in here. So the clinician is making the best therapy decisions they can based on available data for this particular patient. But the data is not necessarily as good as it should be. That physician probably is taking care of more than one patient. And so there’s a lot that’s going on here, even in this simple diagram. Well, our goal initially is to provide support for that patient, to give them guidance in terms of what therapies are going to work in what circumstances.

Now, a step beyond that is to take the physician, at least partly out of the loop and replace that with some syringes here. So the idea here is that we could have a closed loop system. We have a patient in the hospital, it’s hooked up to a feedback system that’s informed by the bio digital twin and of course controlled by what the physician would like to have happen for this patient. Stabilizing pressure, doing whatever is necessary to keep the patient in good shape there. So that is actually one of the things that we’ll be working towards over the next years, is essentially looking to see whether we can improve things in an acute setting, in a hospital setting, where we have lots of information and measurement to work from.

Well, eventually as the model improves and as we have access to better remote sensors, think, you know, advanced versions of a Fitbit or an Apple Watch sort of thing. We can start to move from an acute scenario, where we’re helping someone who’s in trouble right now, to a chronic scenario where people have chronic problems and we’re able to manage them in a way that’s best for them. And then we can finally make it to predicting before something happens. And that will allow us to advance to a point where our focus is simply on wellness of people. So thank you very much.