This paper explores the fallacy of determining conformational energies for polar organic molecules from molecular mechanics force fields. Using Taxol as a test case, it investigates how different force fields can produce downright contradictory results for energetic rankings of Taxol conformations.
The bottom line is simple; do NOT trust energies from force fields. Trust geometries. In case of energies force fields usually overemphasize electrostatic interactions because of lack of explicit solvent representation. Thus sometimes even geometries can be warped because of electrostatics overwhelming the optimization. The one thing force fields are good at calculating on the other hand is sterics.
Running a "complete" conformational search with multiple force fields will usually give you completely different geometries for the global minimum, or at least slightly different ones (depending on the molecule). Thus, trusting the global minimum conformation from any one force field is a big fallacy. Thinking that that global minimum will be the true global minimum in solution is nothing short of blasphemy. And for a bioactive molecule, thinking that the global minimum from a force field search will be the bioactive conformation is just...well, that just means you have been seduced by the dark side of the force field.
Lakdawala, A., Wang, M., Nevins, N., Liotta, D.C., Rusinska-Roszak, D., Lozynski, M., Snyder, J.P. (2001). . BMC Chemical Biology, 1(1), 2. DOI: 10.1186/1472-6769-1-2
So what exactly should one do for a 'realistic' computational conformation search? MM search, then take all low-energy conformers as starting points for DFT optimisation with, e.g. CPCM solvent, followed by a high-level/big basis set single point job, also with solvation, for each conformer?
ReplyDeleteWould you say that your comments apply also to the potential energy functions (if such they are) in protein folding? I'm quite skeptical of potential energy funnels and the like as no one seems to be able to calculate them from first principles. Enlighten me.
ReplyDeleteRetread
Based on my quite limited knowledge I will say this:
ReplyDeleteYes, the same pitfalls that apply to small-molecule FFs would apply to protein pot. fns. Actually FFs for proteins are slightly easier to parametrize because of the limited chemical space of amino acids that they have to be applied to. In case of organic drug-like molecules, chemical space and functionality is vast, making the job much more difficult. Nonetheless, even protein FF will suffer from the common problems of electrostatic overwhelming in the absence of explicit solvent; explicit solvent can give some realistic energies but will be computationally demanding, sort of a catch-22 situation. If you notice it, that's why some of the most successful efforts at predicting protein structure, such as those from David Baker's group at UWash, depend on heuristic, empirical model building and not first-principles approaches. They use libraries of already-existing rotamers to fill in the gaps, with the assumption that nature would have already chosen low-energy structures. But as you know, even then the problem is far from solved because there are so many degenerate motifs for an oligopeptide that have the same energy. The trick is for the protein to explore different degenerate conformations, and then take into account long-range effects in a process that we still don't understand. So I do think that protein FFs have the same problems of predicting energies and that's why right now empirical approaches seem to be the best ones for now. Here's a survey:
http://www.wiley.com/legacy/wileychi/ecc/samples/sample07.pdf
anon: DFT energies for complex, polar molecules can be as fickle as force field energies. but sometimes they correlate well, not with FF energies but with true Boltzmann energies. continuum solvation has its own problems. in general, predicting conformational energies for molecules with any reasonable complexity is still very challenging. for more, see post.
ReplyDeleteThanks. Just what I thought. As you note "there are so many degenerate motifs for an oligopeptide that have the same energy". Yet proteins of biological interest (that we've understand and have crystallized) have a single low energy form (or we couldn't crystallize them). Moreover, they defy the Levinthal paradox and get there quickly without exploring the immense conformational space (3 low energy conformations at each alpha carbon atom) that they could. What fraction of protein space is this? Could we ever find out? See the next chemiotics post. Look at the present one.
ReplyDeleteRetread
Ashutosh - nice article - as always.
ReplyDeleteAlthough different force fields differ in absolute energies - different force fields produce good agreement with reported crystal structures in simulating bound ligand conformations. So all is not lost in the world of simulations based on force fields.