Computational Infrastructure for Multiscale Modelling in Biomedical Engineering

Computational modelling is one of the pillars of scientific research in general and biomedical engineering in particular. Simulating real human body functions and organs on the computer presents us with a host of very favourable features:-Computer simulation is totally non-invasive and therefore 100% safe. At an era where the demand for safer, more effective and more economical treatments is becoming increasingly pressing, computer modelling offers an ideal alternative to all other types of testing (animal, clinical etc.): Where applicable, it allows for a major paradigm shift: all testing takes place on the computer, completely remote from testing on living beings. -Moreover, modelling is perfectly repeatable. This feature is very appealing to scientific research, because it allows for experiments (in this case, virtual experiments, on the computer) to be repeated as many times as necessary, in identical and perfectly controllable conditions: Uncertainties and imperfections, all unavoidable features of physical testing and experimentation, are minimised or eliminated. -These two features lead inevitably to a third characteristic of computational modelling: once the tools are in place, the cost of conducting virtual experiments is minimal or zero. This is in great contract to all other types of testing.The main caveat in computational experimentation of the type discussed is that the models have to be developed first and they need to be comprehensive, accurate and validated as to represent reality. The resources we request in this proposal aim at providing us with equipment that is necessary for the development of such models, as they are very demanding in terms of computational resources. To exemplify, we could mention that in order to simulate blood flow in a diseased artery, a computation that would take about one month on a very powerful personal computer is needed. This time-frame is unacceptable in clinical practice: decisions on disease progress and treatment strategy have to be made within a day or two. This excessive time requirement is equally problematic in the model development and validation stage, i.e. the effort the members of our group get involved in. Model development involves many test computations and rapid turn-around time is essential for evaluating the efficacy of specific techniques, the identification and correction of errors, the improvement of model realism, integration etc. Fortunately, computer technology provides us with an affordable way to increase dramatically the speed of such computations: instead of depending on a single computer for such simulations, we can split the problem at hand into many small tasks and assign each one of those tasks to a separate computer; all those machines communicate with each other over a dedicated fast network and exchange all the information needed. This approach (parallelisation) is the dominating trend in high performance computing nowadays.In the example mentioned before, if the computation that would take one month in a single computer is distributed to, say, 30 identical computers, the time required for the simulation to complete is approximately 1 day (usually it is slightly more because parallelisation is never perfect due to communication overheads), a turn-around time that is perfectly acceptable both for clinical decision-making and for effective algorithm development.We request resources to purchase a computer cluster that will enable us to improve dramatically our simulation methods development cycle and will allow us to conduct research that is more challenging and more useful to our clinical partners. The researchers in the group (mostly doctoral students) shall be able to enhance their productivity and will be able to tackle problems that are currently out of reach. In summary, this proposal will open new roads for us, roads that will allow us to model diseases and treatments that we cannot address right now.