Potential msmbuilder workflow

0-basic.py

# Basic usage / workflow
ds = dataset("**/*.dcd")

dihed = ds.derive('dihed', fmt='dir-npy')
dihed += DihedralFeaturizer().fit_transform(ds)

tica = dihed.derive('tica', fmt='hdf5')
tica += tICA().fit_transform(dihed)

clusters = tica.derive('clusters', fmt='hdf5')
clusters += KMeans().fit_transform(tica)

msm = MSM().fit(clusters)
dump(msm, 'msm.pickl')

1-sample.py

msm = load('msm.pickl')
ds = dataset("**/*.dcd")
inds = msm.sample_from_states()

def structures_from_indices(ds, inds):
  """Maybe implement this as a convenience function in MSMBuilder"""
  trajs = []
  for i, frame in inds:
    trajs += mdtraj.load_frame(ds.meta.loc[i]['traj_fn'], top=ds.meta.loc[i]['top_fn'], frame=frame)
  traj = trajs[0].join(trajs[1:])
  return traj

traj = structures_from_indices(ds, inds)
traj.save("structures.pdb")

2-append_ds.py

# Support adding data to a dataset
ds = dataset("3rvy/*.dcd", top="3rvy.prmtop")
ds += dataset("4lto/*.dcd", top="4lto.prmtop")
print(ds.meta['top_fn'].unique())

# Transform some new data
tica = load('tica.pkl')
tica_trajs = dataset('tica_trajs')
tica_trajs += tica.transform(ds)

3-advanced_meta.py

# support adding custom metadata upon opening a dataset

def get_meta(fn):
  ma = re.match("RUN([0-9]+)/CLONE([0-9]+)", fn)
  n_frames = len(mdtraj.open(fn))
  return {'run': ma.group(1), 'clone': ma.group(2), 'n_frames': n_frames}

ds = dataset("RUN*/CLONE*/concat.xtc", meta=get_meta)
print("Using data from", len(ds.meta['run'].unique()), "runs")

  i like the work flow though i think the learning curve for new comers might be too high.