Computational coding to crack disease!

We've heard how we can grow tissue in the dish in order to study biological processes. But what about taking that information, and creating a computer model?

We're joined now by Katherine Fletcher from Oxford University and Peter Kohl from Imperial College London who are doing precisely this.  Hello to both of you.

Katherine -   Hello.

Peter -   Good evening.

Kate -   We'll be talking to Peter in just a moment about his use of cardiac computer models, but first, Katherine, how do computer models compare with delving into the dish?

Katherine -   Well, if you're asking me whether a model on a screen is better than a model on a dish, I have to say that neither is as good as you would like them to be. Tissue isn't a whole organism and neither is a computational model.  The trick is to make them a useful tool to answer a particular question and where computers are especially good is at integrating lots of different types of information. 

You could imagine at a patient level.  In the case of breast cancer, you might look at an MRI scan, mammogram results, information about genetic markers, and other blood tests.  Maybe the patient's history and risk factors, and if you were able to plug that all into a computer, it would help you sort through the data to find the most interesting points. 

At a more basic science level, we could use computers to integrate different readings from experiments by mapping for example the behaviour that's known about cells onto a geometric mesh representing the anatomy of the organism, to for example, find out what is driving a particular type of irregular behaviour. 

I'm here on behalf of the Virtual Physiological Human Project whose aim is to make computational models of the human body in health and disease so that different models of different parts of the body can be clumped together to answer different questions.

Chris -   Could you just introduce us to the concept of what the Virtual Physiological Human is and how this came about?

Katherine -   Yeah.  It's a European commission funded, very ambitious initiative using computers to integrate data, going from the sort of subcellular or smaller scale levelled up, as well as going back again from the whole human level, back down and mapping them together in ways that are interchangeable.  Since about 2008, the EC has funded 46 projects on a variety of different topics from tumour growth to drug safety, to osteoporosis, and aneurism, each of them working in its own area, but trying to make sure they all use standards so that those models could be used together in different ways in the future.

Chris -   So, they speak a sort of common computer language so that if you're developing a model of the kidney, it will talk to ultimately, the model of the heart, so data from one could inform results to another?

Katherine -   Absolutely!  Doing that in practice is quite tricky.  I mean, just getting software to run on your own computer sometimes, you'll run into difficulties, but that is absolutely the idea.  By using common computer software languages for example, allows you to then plug them together.

Chris -   So, most of the initial setup of the project must've been defining those sorts of standards that the individual groups will work to?

Katherine -   Out of the 46 projects that have been funded, one of them was called the network of excellence whose job was to basically serve as a rallying point for researchers to understand what others in the field are doing, and together, develop best practice in standards.  And so, there have been a lot of technical things developed through that such as ontology's with the standardised vocabulary for describing a model.  If you're interested in this sort of thing, vph-portal.eu is the VPH portal website where you can find all of these information online if you're interested.

Kate -   Why can't we create an entire human model that can work all these out at the same time?

Katherine -   That's actually technically quite complicated to do that.  If you're trying to run a model for example of a human heart beating using a very up-to-date computational model, you have to get that to run not only accurately, but you'd have to get it to run faster than real time because it's no good to predict that a person is going to die of a heart condition in 5 years' time if it takes you 7 years to predict that.  So, you have to get these models to be both accurate and extremely quick before they're useful.

Kate -   I presume when we start the model, we have to plug in all the data we know to get it to work accurately.  Where do we get that data in the first place?

Katherine -   There might be information pulled from scientific literatures, so you might find experiments in papers and you can borrow the data from them.  It might be coming from real world sources so patients.  So for example, MRI data or stands of various body parts that might be diseased.  And the trick is finding ways that you can combine them so that you might be able to map the picture of the MRI of a particular person's tumour state to what it already known also about how other tumours of that type grow, and combining them in a meaningful way to get what treatment this patient might respond to best.

Chris -   So, where are you at now?  So, you have the standards defined, you've got these projects funded to start building models of individual systems.  Where are we in terms of actually having a model person in a computer?

Katherine -   The idea is still not actually to ever run them all as one individual, at least not as a model stand.  The idea is to select the pieces that you're interested in.  So, we have lots of very well developed models for example in the cardiac field, musculoskeletal models are also quite good.  There's still a lot of interesting things done with viruses and also with cancer.  You can imagine all those pieces could all interplay quite well for example.

Kate -   And Peter, you're working on one of those projects to do with that cardiac modelling.  What do you need to look at in order to model how the heart works on a computer?

Peter -   We need to first perhaps start off with agreeing on what we understand then as the terminology of a model.  I think the definition that the dictionaries offer, a simplified representation of reality is very helpful because if you appreciate that any model that you might want to use in whatever circumstance is simplified representation of that reality.  We also realised that any model regardless of whether it is a lung in a dish or a model of the heart will have its own limitations.  If we appreciate that then from there, it follows that the idea and that is coming back to a question you asked slightly earlier, of an all inclusive model of the human is a bit of a contradiction in terms.

Chris -   Are we not really in a position where all that the model knows is all that we know?  So, until we've done the biology, how can we have a computer that's actually any better?

Peter -   If our models did indeed contain all the knowledge we have, that would be great and I don't think we can claim that for any of our model systems including computer models.  If they did have that knowledge however, they would be suitable and useful in different ways.  One would be to treat them as an expert system that can be interrogated their real use for computational model in such as of the heart is to use them to try and understand better the experimental or clinical data that we receive. 

A single cell of the heart is a very complex entity.  These cells are governed by processes at the molecular level that are so complex and that interact via multiple feed forward and feedback mechanisms that certainly my brain is not good enough to interpret findings that we may observe in experimental setting.  So, running computer simulations at the same level and in parallel to experimental studies can be incredibly helpful in understanding what you're doing.  The next step beyond that and that is, I think where it gets really exciting and really interesting is to run simulations first before you go into the lab or before you consider what you might want to do with the patient because you explore the parameter space in theoretical investigations that might allow you to predict what the most likely paths for success might be.  Success in this context depends entirely on the application - that maybe development of a drug or a need for an experimental series that gives you a clear-cut answer.

Kate -   So Peter, when you've been modelling the heart on a computer, what have you found so far by testing that model?

Peter -   One angle that I'm taking is that often experimental researcher who tries to combine experimental and theoretical investigations and these models have been incredibly helpful in planning better experiments.  In particular then, we find that our models do not reproduce what we expect.  That may sound slightly counterintuitive, but if you have the best conceived model there is and it doesn't quite reproduce what you see, then it highlights a shortcoming in our understanding.  It poses a question and science really is all about the question. 

Now in terms of more applied aspects such as development of drugs, I wish we had Gary Mirams here with us who is conducting cardiac single-cell modelling at Oxford University and what Gary has done is to look into how one can try and predict side effects of pharmaceutical agents very early in the development stage.  You see, the most frequent cause for drugs that may otherwise be very useful that requires them to be withdrawn from the market are cardiac side effects.  Now, Gary Miram has developed a technique that is - one might say, reductionist in that he looks at the shape of that electrical signal in a cardiac cell before people look at one single-ion channel type and predict at what might happen.  Now, he uses three ion channel types and that may sound really small, but the impressive thing is that his ability to predict the likelihood of cardiac side effect has improved multiple times over what was possibly before.  It shows that a simplified model has its place to play in the process of explaining phenomena.

Kate -   Thank you so much to Katherine Fletcher from Oxford University and Peter Kohl from Imperial College London.