hbond pmap_mpi output format

[Llauset]

Hello,

Version: 2.0.5

Using pt.hbond and pt.distance work really well for my case study and I would like to use the parallel version pmap_mpi to accelerate de calculation.

The parallelization of pt.distance works straight forward. I'm struggling with the parallelization of pt.hbond.
The outputting format changes with respect to the sequential version.
Sequential: <pytraj.hbonds.DatasetHBond donor_acceptor pairs : 6938>
Parallel: [(OrderedDict([('total_solute_hbonds', array([706, 681, 670, ..., 692, 702, 680], dtype=int32)),
'PHE1472_O-LEU1475_N-H', array([1, 0, 0, ..., 1, 0, 0], dtype=int32)) ....

Trying to obtain the same format I added dtype as follows:
hb_parallelized = pt.pmap_mpi(pt.hbond, traj, distance=3.0, angle=135, dtype="hbond")
And in that case I have the following error:
File "//.conda/envs/openmpi_test_3.7/lib/python3.7/site-packages/pytraj/parallel/base.py", line 51, in concat_hbond
all_keys.update(partial_data[0].keys())
AttributeError: 'DatasetHBond' object has no attribute 'keys'

I tried to work around the problem directly returning the object data_collection and accessing to it data_collection[0][0].get_amber_mask()[0] but doing that I have not all the hbonds.

Can you please tell me if there is a way to change the parallel outputting format to the sequential one or if exist a parameter to obtain directly the same format in the parallel version than in the sequential.

Thank you in advance for your help.

[hainm]

Can you please tell me if there is a way to change the parallel outputting format to the sequential one or if exist a parameter to obtain directly the same format in the parallel version than in the sequential.

Dear @Llauset, unfortunately there is no way to do any thing you mentioned. But we will keep this in mind, I think it's nice to make it work.

For the information: what kind of information you want from pytraj.hbonds.DatasetHBond?

[Llauset]

Dear Hainm,

Thanks for your responsiveness and your answer.
I want to retrieve the list of hydrogen bonds defined by a distance and an angle calculated with pt.hbond and obtained using the amber_mask() function of the pytraj.hbonds.DatasetHBond object. As the trajectory is long I would like to compute ph.hbond in parallel.

If the development of the parallel functionality takes time I would like to try to develop a work around.
Is it possible to rebuild this list 'easily' from the object returned by hb_parallelized = pt.pmap_mpi(pt.hbond, traj, distance=3.0, angle=135)?
How should one proceed? If this is a functionality that may interest you, once implemented, we could share it with you.

[hainm]

I want to retrieve the list of hydrogen bonds

Dear @Llauset: for the parallel version, the returning data is a dict where the keys are total_solute_hbonds and all the hbonds formed during the simulation.

here is an example

In [1]: import pytraj as pt
tra
In [2]: traj = pt.datafiles.load_trpcage()[:]

In [3]: 

In [3]: d = pt.hbond(traj, dtype='dict')

In [4]: d.keys()
Out[4]: odict_keys(['total_solute_hbonds', 'ASN1_O-GLN5_N-H', 'ARG16_O-TRP6_NE1-HE1', 'TYR3_O-LEU7_N-H', 'ILE4_O-LYS8_N-H', 'LEU7_O-GLY10_N-H', 'ASP9_O-SER14_OG-HG', 'SER14_O-ARG16_N-H', 'ASP9_OD2-ARG16_NH1-HH12', 'ASP9_OD2-ARG16_NH2-HH22', 'LEU2_O-TRP6_N-H', 'GLN5_OE1-LYS8_NZ-HZ1', 'ASN1_O-ILE4_N-H', 'TRP6_O-GLY11_N-H', 'SER20_OXT-SER20_OG-HG', 'ASN1_O-TYR3_N-H', 'GLY11_O-SER14_OG-HG', 'ASP9_OD2-ARG16_NE-HE', 'ASN1_OD1-LEU2_N-H', 'ASP9_OD1-LYS8_NZ-HZ1', 'ASP9_OD2-ARG16_NH2-HH21', 'SER20_O-SER20_OG-HG', 'GLY10_O-SER13_N-H', 'GLY10_O-SER13_OG-HG', 'ASP9_OD1-SER14_OG-HG', 'PRO12_O-GLY15_N-H', 'PRO19_O-SER20_OG-HG', 'GLY11_O-SER14_N-H', 'SER13_O-SER13_OG-HG', 'GLN5_O-ASP9_N-H', 'ASP9_OD2-SER14_OG-HG', 'ASP9_OD2-ARG16_NH1-HH11'])

In [5]: d['ASP9_OD2-ARG16_NE-HE']
Out[5]: array([0, 0, 0, ..., 0, 0, 0], dtype=int32)

d['ASP9_OD2-ARG16_NE-HE'] return an array of int with either 0 or 1 value representing the absence or existence of that spefic hbond for specific frame.

Please let me know if that works for you.

[hainm]

Is it possible to rebuild this list 'easily' from the object returned by hb_parallelized = pt.pmap_mpi(pt.hbond, traj, distance=3.0, angle=135)?

So the question is "yes, it's easy"
(d comes from example above)

print(list(set(d) - {"total_solute_hbonds"}))

[hainm]

If this is a functionality that may interest you, once implemented, we could share it with you.

Yes, any contribution to the code is always welcome. Thanks.

[Llauset]

Thank you for your help.

What I did is this function to transform the output of the parallel hbond to the amber_mask and this solve my problem :

def from_hbond_parallel_to_amber_mask(hb_parallelized):
    """
    Convert the keys of hb_parallelized dictionary to amber mask
    :param hb_parallelized: dictionary with the keys of the hydrogen bonds
    :return: list of tuples with the amber mask of the keys
    :rtype: list
    """
    # get all the keys from hb_parallelized dictionary 
    keys = list(hb_parallelized.keys())
    # remove the key 'total_solute_hbonds'
    keys.remove('total_solute_hbonds')
    # change format of keys from HIE4_O-LYS8_NZ-HZ2 to HIE_4@O-LYS_8@NZ-HZ2
    for i in range(len(keys)):
        keys[i] = keys[i].replace("_", " ").replace("-", " ").split()
        # slip the first element after 3 characters
        keys[i][0] = keys[i][0][:3] + '_' + keys[i][0][3:]
        keys[i][2] = keys[i][2][:3] + '_' + keys[i][2][3:]
        acceptor_mask = '@'.join((keys[i][0], keys[i][1]))
        donor_mask = '@'.join((keys[i][2], keys[i][3]))
        keys[i] = '-'.join((acceptor_mask, donor_mask, keys[i][4]))
    # Use function to_amber_mask to convert the keys to amber mask
    amber_masks = list(pt.hbond_analysis.to_amber_mask(keys))
    # split the list of tuples to two independent lists
    distance_masks, angle_masks = list(zip(*amber_masks))
    return distance_masks, angle_masks

[hainm]

thanks @Llauset for the code. Cheers.