Highly parallel parameter estimation of metabolic networks
Partner: Insilico Biotechnology AG
Predictions of metabolic performances through graphical design and evaluation of large metabolic networks are an important part of modern biotechnological processes. This cooperation will combine Insilico's knowledge and technology with the expertise and highly parallel architectures of the HLRS, in order to fully exploit the benefits of modern super computers in this area.
The main aim will be to allow prediction of increases in efficiency based on the total metabolites of a production organism with several hundred components in less than a few days, instead of the several months needed with current methods. This will allow concentrating on more than just small subsystems of metabolism with hardly more then twenty metabolic compounds, as it is still the norm today.
An important tool used for this purpose is the Covariance Matrix Adaptation framework, which has been verified as a valuable help in optimization of various badly conditioned and multi-modal problems in application and science. Over the course of the project it's scalability on multi-processor and multi-node environments will be evaluated and improved through use of hybrid programing models that exploit the architecture of modern supercomputers.
A secondary goal is the utilization of recent advancements in parallel architectures. Especially the rapid development of GPGPUs and the respective frameworks like CUDA or OpenCL show promising possibilities. A careful dissection of the given problem into parts that can be efficiently solved on GPUs may lead to a markable speedup.