Using a meta-GA for parametric optimization of simple GAs in the computational chemistry domain

The determination of the lowest energy conformer for long-chain molecules by exhaustive search methods quickly becomes infeasible as the length increases. Typically, resources required are proportional to the number of possible conformers (shapes), O(3n) where n is the length. A genetic algorithm (GA) that calculates energies in a feasible time is described, using an open-source off-the shelf tool, PyEvolve. By comparing the results using this method with the results from exhaustive search techniques on carnosine, a dipeptide whose energy calculation is currently near the limits of feasibility using exhaustive search methods (n = 8), we obtained quantitative measurements of the performance of this GA. Optimization of a subset of the GAs parameters in a non-adaptive GA was accomplished by encoding the parameters into a genome, and using a meta-GA to tune the algorithm. Our results suggest that PyEvolve's simple GAs with our experimentally-determined parameter values are a computationally feasible method of determining long-chain molecular energies computationally infeasible using other methods.