Alanine dipeptide committor stuff

0_README.md

Alanine dipeptide committor stuff

This directory is part of a collaboration with Max Welling and his colleagues Rianne van den Berg and Thomas Kipf. We're playing with ways to use deep learning neural networks in the context of chemistry.

This is mainly for me to create some useful data for them, and for me to explain some of the tools that I think they will find useful. Some of this may get re-purposed into OpenPathSampling examples.

Here I am taking trajectories from a previously performed (flex-length) TPS calculation on the C7eq to alpha_R states of alanine dipeptide. I select a subset of the snapshots, and run a committor analysis on them. In the end, I give the experiment-by-experiment results, where an "experiment" is the mapping of an initial MD snapshot to the state it ended up in.

• select_snapshots.ipynb: Select the initial set from snapshots. Input requires alanine_dipeptide_tps.nc (calculated for the OPS paper). Output is snapshots.nc.
• commttor_simulation.ipynb: Committor analysis for the selected snapshots. Output is committor_simulation.nc.
• committor_analysis.ipynb: Analysis of the committor simulation. Output is in committor_results.nc. This output includes the mapping of point in phase space to final state (binary value) that we discussed.
• analysis_help.ipynb: How to load the output from the former and get relevant information from it.
• mdtraj_tricks.ipynb: Introduction to a few potentially relevant aspects of MDTraj. Examples of how to create descriptors.

(Rianne and Thomas, the parts that are most relevant to you are mdtraj_tricks.ipynb and analysis_help.ipynb. The rest provides a record of how I'm creating these files.)

---

Information on various programs that interact with each other, to give context for Rianne and Thomas:

• OpenMM: Molecular dynamics code. Unless you're actually doing simulations, you won't need to worry about this.
• MDTraj: Trajectory analysis code, written by some of the same people as OpenMM, but useful for many other packages. You should use this for descriptors.
• OpenMMTools: Useful shortcuts for working with OpenMM. Again, you won't need this unless you actually do simulations.
• OpenPathSampling (OPS): What we use for our path sampling simulations. The files I create are readable by OPS, so you'll be working with some OPS objects.

If you can install OpenPathSampling using conda (you'll need to add the omnia channel), you should get all of these automatically. There are a few things I use that still aren't part of an OPS release, so you might want prefer to use a git-based local install of OPS in practice.

1_select_snapshots.ipynb

2_committor_simulation.ipynb

3_committor_analysis.ipynb

4_analysis_help.ipynb

5_mdtraj_tricks.ipynb