datajoint_to_nwb.py

#!/usr/bin/env python3
import os

import sys
from datetime import datetime
from dateutil.tz import tzlocal
import pytz
import re
import numpy as np
import pandas as pd
import warnings
import tqdm

from pipeline import (reference, subject, acquisition, stimulation, analysis,
                      intracellular, extracellular, behavior, utilities)
import pynwb
from pynwb import NWBFile, NWBHDF5IO

warnings.filterwarnings('ignore', module='pynwb')

# ============================== SET CONSTANTS ==========================================
# Each NWBFile represent a session, thus for every session in acquisition.Session, we build one NWBFile
default_nwb_output_dir = os.path.join('data', 'NWB 2.0')
institution = 'Janelia Research Campus'
hardware_filter = 'Bandpass filtered 300-6K Hz'
related_publications = 'doi:10.1523/JNEUROSCI.3152-17.2018; doi:10.25378/janelia.7489253'
ecephys_fs = 25000

# experiment description and keywords - from the abstract
experiment_description = 'Extracellular electrophysiology recordings with optogenetic perturbations performed on mouse anterior lateral motor cortex (ALM) in delay response task.'
keywords = ['motor planning', 'premotor cortex', 'preparatory activity',
            'short-term memory', 'extracellular electrophysiology', 'intracellular electrophysiology']


def export_to_nwb(session_key, nwb_output_dir=default_nwb_output_dir, save=False, overwrite=True):
    this_session = (acquisition.Session & session_key).fetch1()

    identifier = '_'.join([this_session['subject_id'],
                           this_session['session_time'].strftime('%Y-%m-%d'),
                           this_session['session_id']])

    # =============== General ====================
    # -- NWB file - a NWB2.0 file for each session
    nwbfile = NWBFile(
        session_description=this_session['session_note'],
        identifier=identifier,
        session_start_time=this_session['session_time'],
        file_create_date=datetime.now(tzlocal()),
        experimenter='; '.join((acquisition.Session.Experimenter & session_key).fetch('experimenter')),
        institution=institution,
        experiment_description=experiment_description,
        related_publications=related_publications,
        keywords=keywords)
    # -- subject
    subj = (subject.Subject & session_key).fetch1()
    nwbfile.subject = pynwb.file.Subject(
        subject_id=this_session['subject_id'],
        description=subj['subject_description'],
        genotype=' x '.join((subject.Subject.Allele & session_key).fetch('allele')),
        sex=subj['sex'],
        species=subj['species'])

    # =============== Intracellular ====================
    cell = ((intracellular.Cell & session_key).fetch1()
            if len(intracellular.Cell & session_key) == 1
            else None)
    if cell:
        # metadata
        whole_cell_device = nwbfile.create_device(name=cell['device_name'])
        ic_electrode = nwbfile.create_ic_electrode(
            name=cell['cell_id'],
            device=whole_cell_device,
            description='N/A',
            filtering='N/A',
            location='; '.join([f'{k}: {str(v)}'
                                for k, v in dict((reference.BrainLocation & cell).fetch1(),
                                                 depth=cell['cell_depth']).items()]))
        # acquisition - membrane potential
        mp, mp_wo_spike, mp_start_time, mp_fs = (intracellular.MembranePotential & cell).fetch1(
            'membrane_potential', 'membrane_potential_wo_spike',
            'membrane_potential_start_time', 'membrane_potential_sampling_rate')
        nwbfile.add_acquisition(pynwb.icephys.PatchClampSeries(name='PatchClampSeries',
                                                               electrode=ic_electrode,
                                                               unit='mV',
                                                               conversion=1e-3,
                                                               gain=1.0,
                                                               data=mp,
                                                               starting_time=mp_start_time,
                                                               rate=mp_fs))
        # acquisition - current injection
        if (intracellular.CurrentInjection & cell):
            current_injection, ci_start_time, ci_fs = (intracellular.CurrentInjection & cell).fetch1(
                'current_injection', 'current_injection_start_time', 'current_injection_sampling_rate')
            nwbfile.add_stimulus(pynwb.icephys.CurrentClampStimulusSeries(name='CurrentClampStimulus',
                                                                          electrode=ic_electrode,
                                                                          conversion=1e-9,
                                                                          gain=1.0,
                                                                          data=current_injection,
                                                                          starting_time=ci_start_time,
                                                                          rate=ci_fs))

        # analysis - membrane potential without spike
        mp_rmv_spike = nwbfile.create_processing_module(name='icephys',
                                                        description='Spike removal')
        mp_rmv_spike.add_data_interface(pynwb.icephys.PatchClampSeries(name='icephys',
                                                                       electrode=ic_electrode,
                                                                       unit='mV',
                                                                       conversion=1e-3,
                                                                       gain=1.0,
                                                                       data=mp_wo_spike,
                                                                       starting_time=mp_start_time,
                                                                       rate=mp_fs))

    # =============== Extracellular ====================
    probe_insertion = ((extracellular.ProbeInsertion & session_key).fetch1()
                       if extracellular.ProbeInsertion & session_key
                       else None)
    if probe_insertion:
        probe = nwbfile.create_device(name = probe_insertion['probe_name'])
        electrode_group = nwbfile.create_electrode_group(
            name='; '.join([f'{probe_insertion["probe_name"]}: {str(probe_insertion["channel_counts"])}']),
            description = 'N/A',
            device = probe,
            location = '; '.join([f'{k}: {str(v)}' for k, v in
                                  (reference.BrainLocation & probe_insertion).fetch1().items()]))

        for chn in (reference.Probe.Channel & probe_insertion).fetch(as_dict=True):
            nwbfile.add_electrode(id=chn['channel_id'],
                                  group=electrode_group,
                                  filtering=hardware_filter,
                                  imp=-1.,
                                  x=0.0,  # not available from data
                                  y=0.0,  # not available from data
                                  z=0.0,  # not available from data
                                  location=electrode_group.location)

        # --- unit spike times ---
        nwbfile.add_unit_column(name='sampling_rate', description='Sampling rate of the raw voltage traces (Hz)')
        nwbfile.add_unit_column(name='depth', description='depth this unit (mm)')
        nwbfile.add_unit_column(name='spike_width', description='spike width of this unit (ms)')
        nwbfile.add_unit_column(name='cell_type', description='cell type (e.g. wide width, narrow width spiking)')

        for unit in (extracellular.UnitSpikeTimes & probe_insertion).fetch(as_dict=True):
            # make an electrode table region (which electrode(s) is this unit coming from)
            nwbfile.add_unit(id=unit['unit_id'],
                             electrodes=(unit['channel_id']
                                         if isinstance(unit['channel_id'], np.ndarray) else [unit['channel_id']]),
                             depth=unit['unit_depth'],
                             sampling_rate=ecephys_fs,
                             spike_width=unit['unit_spike_width'],
                             cell_type=unit['unit_cell_type'],
                             spike_times=unit['spike_times'],
                             waveform_mean=unit['spike_waveform'])

    # =============== Behavior ====================
    # Note: for this study, raw behavioral data were not available, only trialized data were provided
    # here, we reconstruct raw behavioral data by concatenation
    trial_seg_setting = (analysis.TrialSegmentationSetting & 'trial_seg_setting=0').fetch1()
    seg_behav_query = (behavior.TrialSegmentedLickTrace * acquisition.TrialSet.Trial
                       * (analysis.RealignedEvent.RealignedEventTime & 'trial_event="trial_start"')
                       & session_key & trial_seg_setting)

    if seg_behav_query:
        behav_acq = pynwb.behavior.BehavioralTimeSeries(name='lick_times')
        nwbfile.add_acquisition(behav_acq)
        seg_behav = pd.DataFrame(seg_behav_query.fetch('start_time', 'realigned_event_time',
                                                       'segmented_lick_left_on',
                                                       'segmented_lick_left_off',
                                                       'segmented_lick_right_on',
                                                       'segmented_lick_right_off')).T
        seg_behav.columns = ['start_time', 'realigned_event_time', 'segmented_lick_left_on',
                             'segmented_lick_left_off', 'segmented_lick_right_on', 'segmented_lick_right_off']
        for behav_name in ['lick_left_on', 'lick_left_off', 'lick_right_on', 'lick_right_off']:
            lick_times = np.hstack(r['segmented_'+behav_name] - r.realigned_event_time + r.start_time
                                   for _, r in seg_behav.iterrows())
            behav_acq.create_timeseries(
                name=behav_name,
                unit='a.u.',
                conversion=1.0,
                data=np.full_like(lick_times, 1),
                timestamps=lick_times)

    # =============== Photostimulation ====================
    photostim = ((stimulation.PhotoStimulation & session_key).fetch1()
                       if stimulation.PhotoStimulation & session_key
                       else None)
    if photostim:
        photostim_device = (stimulation.PhotoStimDevice & photostim).fetch1()
        stim_device = nwbfile.create_device(name=photostim_device['device_name'])
        stim_site = pynwb.ogen.OptogeneticStimulusSite(
            name='-'.join([photostim['hemisphere'], photostim['brain_region']]),
            device=stim_device,
            excitation_lambda=float((stimulation.PhotoStimProtocol & photostim).fetch1('photo_stim_excitation_lambda')),
            location = '; '.join([f'{k}: {str(v)}' for k, v in
                                  (reference.ActionLocation & photostim).fetch1().items()]),
            description=(stimulation.PhotoStimProtocol & photostim).fetch1('photo_stim_notes'))
        nwbfile.add_ogen_site(stim_site)

        if photostim['photostim_timeseries'] is not None:
            nwbfile.add_stimulus(pynwb.ogen.OptogeneticSeries(
                name='_'.join(['photostim_on', photostim['photostim_datetime'].strftime('%Y-%m-%d_%H-%M-%S')]),
                site=stim_site,
                resolution=0.0,
                conversion=1e-3,
                data=photostim['photostim_timeseries'],
                starting_time=photostim['photostim_start_time'],
                rate=photostim['photostim_sampling_rate']))

    # =============== TrialSet ====================
    # NWB 'trial' (of type dynamic table) by default comes with three mandatory attributes:
    #                                                                       'id', 'start_time' and 'stop_time'.
    # Other trial-related information needs to be added in to the trial-table as additional columns (with column name
    # and column description)
    if acquisition.TrialSet & session_key:
        # Get trial descriptors from TrialSet.Trial and TrialStimInfo - remove '_trial' prefix (if any)
        trial_columns = [{'name': tag.replace('trial_', ''),
                          'description': re.search(
                              f'(?<={tag})(.*)#(.*)',
                              str((acquisition.TrialSet.Trial
                                   * stimulation.TrialPhotoStimParam).heading)).groups()[-1].strip()}
                         for tag in (acquisition.TrialSet.Trial * stimulation.TrialPhotoStimParam).heading.names
                         if tag not in (acquisition.TrialSet.Trial
                                        & stimulation.TrialPhotoStimParam).primary_key + ['start_time', 'stop_time']]

        # Trial Events - discard 'trial_start' and 'trial_stop' as we already have start_time and stop_time
        # also add `_time` suffix to all events
        trial_events = set(((acquisition.TrialSet.EventTime & session_key)
                            - [{'trial_event': 'trial_start'}, {'trial_event': 'trial_stop'}]).fetch('trial_event'))
        event_names = [{'name': e + '_time', 'description': d + ' - (s) relative to trial start time'}
                       for e, d in zip(*(reference.ExperimentalEvent & [{'event': k}
                                                                        for k in trial_events]).fetch('event',
                                                                                                      'description'))]
        # Add new table columns to nwb trial-table for trial-label
        for c in trial_columns + event_names:
            nwbfile.add_trial_column(**c)

        photostim_tag_default = {tag: '' for tag in stimulation.TrialPhotoStimParam.heading.names
                                 if tag not in stimulation.TrialPhotoStimParam.primary_key}

        # Add entry to the trial-table
        for trial in (acquisition.TrialSet.Trial & session_key).fetch(as_dict=True):
            events = dict(zip(*(acquisition.TrialSet.EventTime & trial
                                & [{'trial_event': e} for e in trial_events]).fetch('trial_event', 'event_time')))

            trial_tag_value = ({**trial, **events, **(stimulation.TrialPhotoStimParam & trial).fetch1()}
                               if (stimulation.TrialPhotoStimParam & trial)
                               else {**trial, **events, **photostim_tag_default})

            trial_tag_value['id'] = trial_tag_value['trial_id']  # rename 'trial_id' to 'id'
            [trial_tag_value.pop(k) for k in acquisition.TrialSet.Trial.primary_key]

            # convert None to np.nan since nwb fields does not take None
            for k, v in trial_tag_value.items():
                trial_tag_value[k] = v if v is not None else np.nan

            trial_tag_value['delay_duration'] = float(trial_tag_value['delay_duration'])  # convert Decimal to float

            # Final tweaks: i) add '_time' suffix and ii) remove 'trial_' prefix
            events = {k + '_time': trial_tag_value.pop(k) for k in events}
            trial_attrs = {k.replace('trial_', ''): trial_tag_value.pop(k)
                           for k in [n for n in trial_tag_value if n.startswith('trial_')]}

            nwbfile.add_trial(**trial_tag_value, **events, **trial_attrs)

    # =============== Write NWB 2.0 file ===============
    if save:
        save_file_name = ''.join([nwbfile.identifier, '.nwb'])
        if not os.path.exists(nwb_output_dir):
            os.makedirs(nwb_output_dir)
        if not overwrite and os.path.exists(os.path.join(nwb_output_dir, save_file_name)):
            return nwbfile
        with NWBHDF5IO(os.path.join(nwb_output_dir, save_file_name), mode = 'w') as io:
            io.write(nwbfile)
            print(f'Write NWB 2.0 file: {save_file_name}')

    return nwbfile


# ============================== EXPORT ALL ==========================================

if __name__ == '__main__':
    if len(sys.argv) > 1:
        nwb_outdir = sys.argv[1]
    else:
        nwb_outdir = default_nwb_output_dir

    for skey in acquisition.Session.fetch('KEY'):
        export_to_nwb(skey, nwb_output_dir=nwb_outdir, save=True)