Proteins are involved in almost all processes that take part in cells of living organisms. Understanding of their function is important for the understanding of biological processes. Correlated mutations in proteins, which take place over the course of their evolution, are closely related to their structure and function. Several different algorithms for detection of correlated mutations have been developed, all very computationally intensive. One of such algorithms is the OMES algorithm, which is based on the statistical χ²-test and random shuffling of protein sequences. The aim of the thesis is paralelisation and implementation of the OMES algorithm on graphics processing units using the CUDA platform. Graphics processing units are specialized processors, which nowadays probably offer the best computing power to price ratio. Compared to the implementation on a central processing unit we achieved a 100-fold speedup and used less energy.
|
