BIASING A MONTE-CARLO CHAIN GROWTH METHOD WITH RAMACHANDRANS PLOT - APPLICATION TO 20-L-ALANINE

In this paper, we explore the possibility of using experimental observations in the Monte Carlo chain growth method that we have previously developed. In this method, the macromolecule (peptide, protein, nucleic acid, etc.) is grown atom-by-atom (or residue-by-residue, etc.) and partial chains are replicated according to their Boltzmann weights. Once the molecule completed, we are left with a Boltzmann-distributed ensemble of configurations. For long molecules, an efficient sampling of the (extremely large) phase space is difficult for obvious reasons (existence of many local minima, limited computer memory, etc.). In the case in which one is mainly interested in the low energy conformations, we have incorporated in the growth scheme experimental observations taken from the Protein Data Banks. More precisely, we have considered the case of twenty-L-alanine and we have used the (experimental) Ramachandran's plot for this residue. The biased growth procedure as follows: (a) each time one adds along the main backbone chain, either a carbon atom belonging to a carbonyl group, or a nitrogen atom, its dihedral angle (phi) or (psi) is drawn with a probability law that reflects the experimental Ramachandran (phi, psi) plot; (b) the bias introduced in this way is canceled through an extra term in the energy (replication energy = true energy + bias energy); (c) the configurations, generated at T = 1000 K, are then energy minimized. We have worked with an all-atom CHARMM force field, and Ramachandran's plot for the alanine was modeled through three angular zones (alpha-helix, beta-sheet, coil). In our calculations, the probabilities of the alpha(p(alpha)) and beta(P-beta) regions have been varied in large proportions (p(alpha) between 0.64 and 0.19, the ''experimental'' value being 0.59). The results, based on 35 ''unbiased'' and 25 ''biased'' (or ''guided'') distinct minimized configurations clearly demonstrate the efficiency of the method. The low energy configurations, for all tested values of p alpha, have a total (or almost total) alpha helix content. The unbiased configurations have much higher energies (in general, even higher than the left-handed helix). Note that the method is not ''alpha helix in-alpha helix out,'' since working at T = 300 K with the experimental (p alpha = 0.59) value yields configurations partially frozen in a C-7(eq) alanine dipeptide type of local minima. (C) 1993 John Wiley and Sons, Inc.