yank
yank copied to clipboard
choderalab
Incorporate QM into ligand parameterization pipeline
As a YANK newbie, it seems the existing ligand parameterization pipeline doesn't use any ab initio QM calculations.
Since Schrodinger runs their force field builder and Jaguar to fill in missing torsions from OPLS3e before every FEP calculation, is there any way to mimic this behavior in YANK? The closest turn-key solutions I've been able to find for this problem are atb, gaamp, and fftk. From what I can tell, it seems there was some talk in OpenMM about CGenFF support a while ago. The existing YANK benchmarks also don't seem to use any RESP charges.
Hi @tantrev . You are right. Currently, the YAML interface doesn't support fitting torsions or passing the charge density as an input to antechamber so you can't use it to generate RESP charges.
For RESP, the easiest way to incorporate them would be to generate the mol2 file with RESP charges yourself into a mol2 and then use that mol2 file as an input for YANK. For fitting torsions, I'm afraid you'll have to skip the pipeline, fit the torsions separately, and provide topology/coordinate files for the final system yourself (see syntax here: http://getyank.org/latest/tutorial.html#when-the-automatic-pipeline-won-t-do).
Hi @andrrizzi . Thank you for the kind reply and explanation - I appreciate the clarification. It sounds like I will be crossing my fingers for future QM integration into YANK/Perses. :)
For what it's worth, the CHARMM ecosystem seems to have the most robust automated torsion-fitting tools with applications like QuickFF available, but it seems that GAFF has become the standard in the YANK/Perses ecosystem. GAAMP may be the best solution for being easily "bolted on" to the existing infrastructure here, though I am a little cautious about its automated fitting. Perhaps AMBER's PARAMFIT may be the best long-term solution for filling in missing torsions (and using AMBER's other tools for obtaining RESP charges), but it seems like the protocol has a little ways to go before being easily "turn-key". Long-term, I suspect there may be some desire to to use the free orca for ab initio calculations instead of Gaussian, but at least many academics still have access to Gaussian.
If it is of any help, I've just merged this PR (#1207) that adds support in YAML to read CHARMM psf files so you may go through their setup pipeline and (rather than the AMBER-based one integrated in YANK) just feed the input psf + pdb to YANK.
That's great, thank you @andrrizzi (and @jeffcomer )! I'll probably give it a try in the next couple weeks as I play around with different pipelines.
I was looking more into potential QM parameterization, and while it seems that there really isn't any existing free/open tool like Schrodinger's FFBuilder for parameterizing force fields using multiple molecules simultaneously (as would be needed for relative FEP calculations), the single-molecule approach could potentially be used for single-molecule absolute energy FEP calculations. Might be useful for hit enrichment anyways (as a side note, Schrodinger just added such calculations to their latest release).
I was wrong about the previous pipelines all lacking support for multiple-molecules. For CHARMM, the parameterization seems like it can be thrown into a database - at least according to SILCSBio's page on the matter (who also seems to have the easiest turn-key tool for scaling up CHARMM parameterization).
If that's the case, it would be really awesome if perses could also get the CHARMM psf support. 😄
Just thought I'd add an update that FFParam seems to be the best turn-key solution for CHARMM right now.
In another update, I've come across "parameterize" from Accelera that is more command-line friendly (FFParam is heavily GUI based).