Hybrid Metaheuristic Approach for Nonlocal Optimization of Molecular Systems

Abstract

Accurate modeling of molecular systems requires a good knowledge of the structure; therefore, conformation searching/optimization is a routine necessity in computational chemistry. Here we present a hybrid metaheuristic optimization (HMO) algorithm, which combines ant colony optimization (ACO) and particle swarm optimization (PSO) for the optimization of molecular systems. The HMO implementation meta-optimizes the parameters of the ACO algorithm on-the-fly by the coupled PSO algorithm. The ACO parameters were optimized on a set of small difluorinated polyenes where the parameters exhibited small variance as the size of the molecule increased. The HMO algorithm was validated by searching for the closed form of around 100 molecular balances. Compared to the gradient-based optimized molecular balance structures, the HMO algorithm was able to find low-energy conformations with a 87% success rate. Finally, the computational effort for generating low-energy conformation(s) for the phenylalanyl-glycyl-glycine tripeptide was approximately 60 CPU hours with the ACO algorithm, in comparison to 4 CPU years required for an exhaustive brute-force calculation.

Memetic algorithms for ligand expulsion from protein cavities

“…The representation of the memory is equivalent to pheromones trails in the Ant Colony Optimization (ACO) algorithms 9 . The concept of ACO has been exploited recently in the computational chemistry context 10,41 . In MERA, every ligand leaves the pheromone trail during simulation and that trail is used as information for the next ligands positions on the dissociation pathway.…”

Protein Structure Optimization with a "Lamarckian"' Ant Colony Algorithm

“…These will be referred to as BLN-46 and BLN-69 in the rest of this manuscript. The largest molecule previously investigated with ACO was a peptide with 10 rotatable torsion angles [18]. The BLN model proteins have 43 and 66 rotatable torsions and, therefore, have substantially larger conformational spaces.…”

“…Using a restart operator ensures that, eventually, all searches locate the global minimum. In ACO searches on other system [18], [32], the efficiency of the search is improved by including the best solution found so far in the trail update. We include a global best update in the LamarckiAnt algorithm, with a fraction, g, of the trail update in each cycle supplied by the best solution found so far.…”

“…One search algorithm that has only seen limited applications in molecular structure optimization [15], [16], [17], [18] is ant colony optimization (ACO) [19]. This algorithm is inspired by the foraging of colonies of ants, which tend to find the shortest path to a source of food in spite of the fact that individual ants have no knowledge of the overall landscape.…”

“…To account for the continuous nature of the energy function, pheromone was laid down in the chosen bin, with smaller amounts deposited in neighboring bins. Dresselhaus et al [18] modified this approach by using particle-swarm optimization to optimize the parameters used in the ACO search.…”