Reweighting with ai2-kit#

Introduction#

TODO

Example#

Given we have the following data:

traj.lammpstrj: a LAMMPS trajectory file
COLVAR: a PLUMED COLVAR file, the COLVAR file should align with the trajectory file
dp-baseline.pb: a DeepMD baseline model
dp-target.pb: a DeepMD target model

Sampling 2000 frames from the trajectory and COLVAR file#

We don’t need to use the full trajectory for reweighting, so here we will use ai2-kit to sampling 2000 frames from the trajectory. As the beginning frame may not be the best frame to start with, so we will also skip the first 100000 frames.

Note that we need to ensure the frames in the COLVAR file are aligned with the trajectory file, so we need to use the same random seed for both sampling.

ai2-kit tool frame read traj.lammpstrj --rp 'TIMESTEP' - slice 100000: - sample 2000 --method random --seed 10 - write 2000.lammpstrj 
ai2-kit tool ase read 2000.lammpstrj --specorder [Ag,O] - to_dpdata - write 2000-dpdata

ai2-kit tool frame read COLVAR --frame_size 1 --header_size 1 - slice 100000: - sample 2000 --method random --seed 10 - write 2000-colvar --keep_header

The reason of not using ase tool to sample is ase will take extra time to parse the trajectory file, by using frame tool to sample and then use ase tool to convert the data to dpdata format will save time.

Here we use the same slice and sample parameters for both traj.lammpstrj and COLVAR to ensure they are aligned.

Use baseline and target model to calculate the energy#

Now we need to use both the baseline and target model to calculate the energy of the sampled frames.

ai2-kit tool dpdata read 2000-dpdata/* --nolabel - eval dp-baseline.pb - write 2000-baseline
ai2-kit tool dpdata read 2000-dpdata/* --nolabel - eval dp-target.pb - write 2000-target

Note that we use --nolabel to ignore label in the dpdata file.

Calculate reweighting FES#

Now that we have all the data we need, we can calculate the reweighting FES.

 ai2-kit algorithm reweighting load_energy 2000-baseline/**/energy.npy --tag baseline  - load_energy 2000-target/**/energy.npy --tag target \
   - load_colvar 2000-COLVAR \
   - reweighting --cv d1 --bias opes.bias --temp 800 --save_fig_to fes.png --save_json_to result.json

The above command will calculate the reweighting FES and save the FES to fes.png and reweighting result to result.json.