This git repo contains all the data + ipython notebooks necessary to reproduce the results in Sultan & Pande paper(JCTC 2017). The overarching idea is that we can use linear and non-linear kernel tica coordinates to enhance the sampling of regular MD trajectories via Metadyanmics. The nice thing being that we are dropping gaussians along the slowest set of CVs. In addition, with the landmark kernel extension we are actually dropping guassians in full phase space if we use the RMSD distance metric. The data can then be used for qualitivative inspection of trajectories or combined into the MSM framework via TRAM/WHAM/MBAR etc. We also show that tica metadynamics can be extended to multiple orthogonoal coordinates by combining it a replica scheme(Bias-Exchange).
Our numerical experiments have also shown that tica metandynamics can reversibly drive transitions along the slowest modes even when NO such transition was seen in the tica training data. However, this does require information about the high and low free energy states and regular MD sampling to be performed in those basins.
The paper is currently under second round of reviews, and since ACS is not amenable to pre-prints, this repo doesn't contain the actual paper document/pdf.
The repo is organized into two folder sections.
- Alanine
- BPTI
Both sections contain a model(or several model folders),and a sub folder metad_data which contains the results from running the metadynamics trajectories.The sections also contain a ipython notebook(in the ipynb folder) which contains all the scripts for making the figures as well as generating the plumed input files. We are currently creating a package that can hopefully automate the latter portion.
Reproducing trajectories:
To reproduce the enhanced sampling trajectories simply run python simulate.py
in the appropriate metad_datafolder.The plumed.py file contains the
actual plumed script that is added as an external force to OpenMM.In the case of BPTI,
follow the instructions of the MPI connected replica exchange code code.
Remaking Figures and re-generating plumed files:
The ipython notebooks contain all the necessary data scripts to make the figures shown in the paper.If
you regenerate the trajectories then you will need to re-run plumed sum_hills --hills HILLS
to integrate the FES to obtain a new fes.dat file.
Notes:
-
For BPTI, we are unable to provide the raw trajectories, featurizer objects, and landmark pdbs due to copy right issues. We note that we used the following 0-indexed locations from 1 ms trajectory(sub-sampled to 25ns) to generate the model. [22337, 37384, 21079, 18818, 34032, 8046, 33230, 8768, 33659, 21386, 40896, 6583, 24395, 1703, 33804, 24607, 8404, 33517, 31602, 32847, 32165, 33077, 14153, 14657, 37158, 38741, 32583, 23406, 35820, 28077, 28222, 17046, 2831, 9374, 33830, 7085, 36247, 29170, 32350, 33175, 18714, 32962, 25604, 33886, 37648, 22345, 28227, 5104, 37812, 32785]
-
The simulations are perfomed using openmm and plumed via the openmm plumed plugin.
-
Please ensure that you install plumed with libmatheval support.