Speeding up a genetic algorithm for EPR-based spin label characterization of biosystem complexity

Complexity of biological systems is one of the toughest problems for any experimental technique. Complex biochemical composition and a variety of biophysical interactions governing the evolution of a state of a biological system imply that the experimental response of the system would be superimposed of many different responses. To obtain a reliable characterization of such a system based on spin-label Electron Paramagnetic Resonance (EPR) spectroscopy, multiple Hybrid Evolutionary Optimization (HEO) combined with spectral simulation can be applied. Implemented as the GHOST algorithm this approach is capable of handling the huge solution space and provides an insight into the "quasicontinuous© distribution of parameters that describe the biophysical properties of an experimental system. However, the analysis procedure requires several hundreds of runs of the evolutionary optimization routine making this algorithm extremely computationally demanding. As only the best parameter sets from each run are assumed to contribute into the final solution, this algorithm appears far from being optimized. The goal of this study is to modify the optimization routine in a way that 20-40 runs would be enough to obtain qualitatively the same characterization. However, to keep the solution diversity throughout the HEO run, fitness sharing and newly developed shaking mechanisms are applied and tested on various test EPR spectra. In addition, other evolutionary optimization parameters such as population size and probability of genetic operators were also varied to tune the algorithm. According to the testing examples a speed-up factor of 5-7 was achieved