| Regional Biophysics Meeting 2005, March 16-20, Zreče, Slovenia | [ComputModel] |
In order to obtain reliable characterization of complex biological system based on spin-labeling EPR experiment, multiple Hybrid Evolutionary Optimizations (HEO) (repeated 200 times) combined with spectral simulations can be applied. Large solution-space detection enables us to get some insight into “quasi-continuous” distribution of parameters that describe the biophysical properties of the complex system. Until recently only the best parameter set has contributed from each run to the final solution to filtering according to the corresponding value of Hi2 and local solution density as well as finally to be presented with the GHOST condensation algorithm. Since this procedure implies 200 runs of the optimization routine, it comes out as a powerful but very slow and computational demanding approach. The goal of this study is therefore to modify the optimization routine in a way that for example 20-40 runs will be enough to obtain the same information with GHOST condensation approach. Some implementation like fitness sharing, and newly developed shaking have been applied for testing purpose on various test spectra that represent the large range of possible real applications. In addition, other parameters typical for evolutionary optimization like population size, various probabilities, shaking intensities, etc., were varied to optimize the approach. The evaluations of the modifications were done according to the minimal fitness achieved in a single run and in all 20 runs, ability of keeping solution diversity within a single run on continuous-like problems, GHOST qualities, run contribution histograms, and detected slicing densities. According to the testing examples a gain of a speed-up factor in the range from 5 to 10 was achieved.
Email: janez.strancar@ijs.si
Address: Laboratory of biophysics - EPR center,, Jozef Stefan Institute,, Jamova 39,, SI-1000 Ljubljana, Slovenia