def save_dataset_instance(db_filename, kw_filename, instance_filename):
# Create a new Dataset instance
dataset = Dataset('./raw_data/' + db_filename + '.txt')
# Add some features
dataset.add_features('./raw_data/' + kw_filename + '.txt')
# Save new file
dataset.save('./raw_data/' + instance_filename + '.pkl')
return dataset
def __init__(self,
db,
dataset=None,
studies=None,
features=None,
reset_db=False,
reset_dataset=False,
download_data=False):
"""
Initialize instance from a pickled Neurosynth Dataset instance or a
pair of study and analysis .txt files.
Args:
db: the SQLAlchemy database connection to use.
dataset: an optional filename of a pickled neurosynth Dataset
instance.
studies: name of file containing activation data. If passed, a new
Dataset instance will be constructed.
features: name of file containing feature data.
reset_db: if True, will drop and re-create all database tables
before adding new content. If False (default), will add content
incrementally.
reset_dataset: if True, will regenerate the pickled Neurosynth
dataset.
download_data: if True, ignores any existing files and downloads
the latest Neurosynth data files from GitHub.
"""
if (studies is not None and not os.path.exists(studies)) \
or settings.RESET_ASSETS:
print("WARNING: RESETTING ALL NEUROSYNTH ASSETS!")
self.reset_assets(download_data)
# Load or create Neurosynth Dataset instance
if dataset is None or reset_dataset or (isinstance(dataset, str) and
not os.path.exists(dataset)):
print("\tInitializing a new Dataset...")
if (studies is None) or (features is None):
raise ValueError(
"To generate a new Dataset instance, both studies and "
"analyses must be provided.")
dataset = Dataset(studies)
dataset.add_features(features)
dataset.save(settings.PICKLE_DATABASE)
else:
print("Loading existing Dataset...")
dataset = Dataset.load(dataset)
if features is not None:
dataset.add_features(features)
self.dataset = dataset
self.db = db
if reset_db:
print("WARNING: RESETTING DATABASE!!!")
self.reset_database()
def __init__(self, db, dataset=None, studies=None, features=None,
reset_db=False, reset_dataset=False, download_data=True):
"""
Initialize instance from a pickled Neurosynth Dataset instance or a
pair of study and analysis .txt files.
Args:
db: the SQLAlchemy database connection to use.
dataset: an optional filename of a pickled neurosynth Dataset
instance.
Note that the Dataset must contain the list of Mappables (i.e.,
save() must have been called with keep_mappables set to
True).
studies: name of file containing activation data. If passed, a new
Dataset instance will be constructed.
features: name of file containing feature data.
reset_db: if True, will drop and re-create all database tables
before adding new content. If False (default), will add content
incrementally.
reset_dataset: if True, will regenerate the pickled Neurosynth
dataset.
download_data: if True, ignores any existing files and downloads
the latest Neurosynth data files from GitHub.
"""
if (studies is not None and not os.path.exists(studies)) \
or settings.RESET_ASSETS:
print "WARNING: RESETTING ALL NEUROSYNTH ASSETS!"
self.reset_assets(download_data)
# Load or create Neurosynth Dataset instance
if dataset is None or reset_dataset or (isinstance(dataset, basestring) and not os.path.exists(dataset)):
print "\tInitializing a new Dataset..."
if (studies is None) or (features is None):
raise ValueError(
"To generate a new Dataset instance, both studies and "
"analyses must be provided.")
dataset = Dataset(studies)
dataset.add_features(features)
dataset.save(settings.PICKLE_DATABASE, keep_mappables=True)
else:
print "\tLoading existing Dataset..."
dataset = Dataset.load(dataset)
if features is not None:
dataset.add_features(features)
self.dataset = dataset
self.db = db
if reset_db:
print "WARNING: RESETTING DATABASE!!!"
self.reset_database()
def _getdata():
"""Downloads data from neurosynth and returns it as a Dataset.
Also pickles the dataset for future use."""
LOG.warning("Downloading and processing Neurosynth database")
os.makedirs("data", exist_ok=True)
from neurosynth.base.dataset import download
download(path="data", unpack=True)
data = Dataset("data/database.txt")
data.add_features("data/features.txt")
data.save("data/dataset.pkl")
return data
def fetch_neurosynth_dataset(data_dir, return_pkl=True):
"""Downloads the Neurosynth dataset
Parameters
----------
data_dir : str
Directory in which to download the dataset.
return_pkl : bool
If true, creates and returns the .pkl file. Otherwise returns
the dataset and features files.
Returns
-------
tuple, str
If save_pkl is false, returns a tuple containing the path to the
database.txt and the features.txt file. Otherwise returns the path
to the .pkl file.
"""
if not os.path.isdir(data_dir):
os.mkdir(data_dir)
dataset_file = os.path.join(data_dir, "database.txt")
if not os.path.isfile(dataset_file):
logging.info("Downloading the Neurosynth dataset.")
download(data_dir, unpack=True)
feature_file = os.path.join(data_dir, "features.txt")
if return_pkl:
pkl_file = os.path.join(data_dir, "dataset.pkl")
if not os.path.isfile(pkl_file):
logging.info(
"Converting Neurosynth data to a .pkl file. This may take a while."
)
dataset = Dataset(dataset_file, feature_file)
dataset.save(pkl_file)
return pkl_file
return (dataset_file, feature_file)
# <codecell>
# Create a new Dataset instance
dataset = Dataset('data/database.txt')
# Add some features
dataset.add_features('data/features.txt')
# <markdowncell>
# Because this takes a while, we'll save our Dataset object to disk. That way, the next time we want to use it, we won't have to sit through the whole creation operation again:
# <codecell>
dataset.save('dataset.pkl')
# <markdowncell>
# Now in future, instead of waiting, we could just load the dataset from file:
# <codecell>
dataset = Dataset.load('dataset.pkl') # Note the capital D in the second Dataset--load() is a class method
# <markdowncell>
# ## Doing stuff with Neurosynth
# Now that our Dataset has both activation data and some features, we're ready to start doing some analyses! By design, Neurosynth focuses on facilitating simple, fast, and modestly useful analyses. This means you probably won't break any new ground using Neurosynth, but you should be able to supplement results you've generated using other approaches with a bunch of nifty analyses that take just 2 - 3 lines of code.
#
# ### Simple feature-based meta-analyses
def create_dataset(database_location, feature_location):
dataset = Dataset(database_location)
dataset.add_features(feature_location)
dataset.save('neurosynth-dataset.pkl')
return dataset
xfm2vol.run()
#make masks to input into neurosynth
def cluster2masks(clusterfile):
clustermap = nb.load(clusterfile).get_data()
for x in range(1, clustermap.max() + 1):
clustermask = (clustermap == x).astype(int)
nImg = nb.Nifti1Image(clustermask, None)
nb.save(
nImg,
os.path.abspath(clusterfile + '_clustermask' + str(x) + '.nii'))
cluster2masks(volume_file)
dataset_file = '/home/raid3/watanabe/neurosynth/data/dataset.pkl'
if not os.path.exists(dataset_file):
dataset = Dataset('/home/raid3/watanabe/neurosynth/data/database.txt')
dataset.add_features('/home/raid3/watanabe/neurosynth/data/features.txt')
dataset.save(dataset_file)
else:
dataset = cPickle.load(open(dataset_file, 'rb'))
clustermask = volume_file + '_clustermask' + str(3) + '.nii'
ids = dataset.get_ids_by_mask(clustermask)
features = dataset.feature_table.get_features_by_ids(ids)
#mri_surf2vol --identity fsaverage4 --surfval /scr/ilz1/Data/attemptsurface.nii --hemi 'lh' --o /scr/ilz1/Data/results/surf2volume.nii --template /scr/ilz1/Data/freesurfer/fsaverage4/mri/orig.mgz
def create_dataset(database_location, feature_location):
dataset = Dataset(database_location)
dataset.add_features(feature_location)
dataset.save('dataset-old.pkl')
print 'created dataset'
return dataset
class Neurosynth:
def __init__(self,
datadir='../data/neurosynth',
verbose=True,
ma_count_thresh=16,
meta_image='consistency_z',
resolution=3):
self.dataset = None
self.concepts = None
self.concepts_df = None
self.concept_pmids = {}
self.datadir = datadir
self.datafile = os.path.join(datadir, 'database.txt')
self.verbose = verbose
self.ma_count_thresh = ma_count_thresh
self.meta_image = meta_image
self.resolution = resolution
self.imagedir_resampled = None
self.image_concepts = None
self.desmtx = None
if not os.path.exists(os.path.join(self.datadir, 'database.txt')):
print('downloading neurosynth data')
ns.dataset.download(path='/tmp', unpack=True)
print('extracting data')
tfile = tarfile.open("/tmp/current_data.tar.gz", 'r:gz')
if not os.path.exists(self.datadir):
os.mkdir(self.datadir)
tfile.extractall(self.datadir)
os.remove("/tmp/current_data.tar.gz")
print('done creating dataset in', self.datadir)
self.imagedir = os.path.join(self.datadir, 'ma_images')
if not os.path.exists(self.imagedir):
os.mkdir(self.imagedir)
def get_dataset(self, force_load=False):
if os.path.exists(os.path.join(self.datadir,
'dataset.pkl')) and not force_load:
print('loading database from',
os.path.join(self.datadir, 'dataset.pkl'))
self.dataset = Dataset.load(
os.path.join(self.datadir, 'dataset.pkl'))
else:
print('loading database - this takes a few minutes')
self.dataset = Dataset(os.path.join(self.datadir, 'database.txt'))
self.dataset.add_features(
os.path.join(self.datadir, 'features.txt'))
self.dataset.save(os.path.join(self.datadir, 'dataset.pkl'))
def get_concepts(self, force_load=False):
if os.path.exists(os.path.join(self.datadir,
'concepts_df.csv')) and not force_load:
print('using cached cognitive atlas concepts')
self.concepts_df = pandas.read_csv(
os.path.join(self.datadir, 'concepts_df.csv'))
else:
self.concepts_df = get_concept().pandas
self.concepts_df.to_csv(
os.path.join(self.datadir, 'concepts_df.csv'))
self.concepts = self.concepts_df.name.tolist()
def get_concept_pmids(self, retmax=2000000, force_load=False):
# get the pmids for each concept that are in neurosynth
# for single-word concepts we use the neurosynth search tool
# for phrases we use pubmed
if os.path.exists(os.path.join(
self.datadir, 'concept_pmids.pkl')) and not force_load:
print('using cached concept_pmids')
self.concept_pmids = pickle.load(
open(os.path.join(self.datadir, 'concept_pmids.pkl'), 'rb'))
return
print('loading all neurosynth pmids')
all_neurosynth_ids = self.dataset.image_table.ids.tolist()
for id in self.concepts:
time.sleep(0.5)
handle = Entrez.esearch(db="pubmed",
retmax=retmax,
term='"%s"' % id)
record = Entrez.read(handle)
handle.close()
# make sure we got all the records - rerun if we didn't
if int(record['Count']) > retmax:
handle = Entrez.esearch(db="pubmed",
retmax=int(record['Count']),
term='"%s"' % id)
record = Entrez.read(handle)
handle.close()
records_int = [int(i) for i in record['IdList']]
ns_pmids = intersect(all_neurosynth_ids, records_int)
print('pubmed found', len(ns_pmids), 'matching pmids for', id)
self.concept_pmids[id] = ns_pmids
pickle.dump(
self.concept_pmids,
open(os.path.join(self.datadir, 'concept_pmids.pkl'), 'wb'))
def get_concept_images(self, force_load=False):
for c in self.concept_pmids.keys():
if not force_load and os.path.exists(
os.path.join(
self.imagedir,
'%s_specificity_z.nii.gz' % c.replace(' ', '-'))):
continue
if len(self.concept_pmids[c]) < self.ma_count_thresh:
#print('skipping',c,len(self.concept_pmids[c]),'pmids')
continue
print('running meta-analysis for', c)
ma = meta.MetaAnalysis(self.dataset, self.concept_pmids[c])
ma.save_results(self.imagedir, c.replace(' ', '-'))
if force_load or not os.path.exists(
os.path.join(self.imagedir, 'mask_image.nii.gz')):
# make mask of voxels with zero standard deviation
concept_images = glob.glob(
os.path.join(self.imagedir, '*_%s.nii.gz' % self.meta_image))
imgdata = numpy.zeros((91, 109, 91, len(concept_images)))
print('loading concept images to compute std')
for i, c in enumerate(concept_images):
tmp = nibabel.load(c).get_data()
imgdata[:, :, :, i] = tmp
imgstd = numpy.std(imgdata, axis=3)
maskdata = (imgstd > 0).astype('int')
maskimg = nibabel.Nifti1Image(maskdata,
affine=nibabel.load(c).affine)
maskimg.to_filename(
os.path.join(self.imagedir, 'mask_image.nii.gz'))
def get_resampled_images(self, shape=None, affine=None, force_load=False):
# use 3 mm as default
if not shape:
shape = [60, 72, 60]
affine = numpy.array([[-3, 0, 0, 90], [0, 3, 0, -126],
[0, 0, 3, -72], [0, 0, 0, 1]])
self.resolution = affine[1, 1].astype('int')
print('resampling data to %d mm' % self.resolution)
self.imagedir_resampled = os.path.join(
self.datadir, 'ma_images_%dmm' % self.resolution)
if not os.path.exists(self.imagedir_resampled):
os.mkdir(self.imagedir_resampled)
concept_images = glob.glob(
os.path.join(self.imagedir, '*_%s.nii.gz' % self.meta_image))
for c in concept_images:
if force_load or not os.path.exists(
os.path.join(self.imagedir_resampled,
os.path.basename(c))):
img = nilearn.image.resample_img(c,
target_affine=affine,
target_shape=shape)
img.to_filename(
os.path.join(self.imagedir_resampled, os.path.basename(c)))
if not os.path.exists(
os.path.join(self.datadir,
'mask_%dmm.nii.gz' % self.resolution)):
# make MNI mask at chosen resolution
mask = os.path.join(
os.environ['FSLDIR'],
'data/standard/MNI152_T1_2mm_brain_mask.nii.gz')
maskimg = nilearn.image.resample_img(mask,
target_affine=affine,
target_shape=shape)
maskimg.to_filename(
os.path.join(self.datadir,
'mask_%dmm.nii.gz' % self.resolution))
def load_concept_images(self, force_load=True):
concept_images = glob.glob(
os.path.join(self.imagedir_resampled,
'*_%s.nii.gz' % self.meta_image))
concept_images.sort()
self.image_concepts = [
os.path.basename(i).split('_')[0] for i in concept_images
]
if os.path.exists(
os.path.join(self.datadir,
'imgdata_%dmm.npy' % self.resolution)):
self.imgdata = numpy.load(
os.path.join(self.datadir,
'imgdata_%dmm.npy' % self.resolution))
# make sure it's the right size
if self.imgdata.shape[1] == len(concept_images):
print('using cached concept image data')
return
masker = nilearn.input_data.NiftiMasker(
mask_img=os.path.join(self.datadir,
'mask_%dmm.nii.gz' % self.resolution),
target_shape=[60, 72, 60],
target_affine=numpy.array([[-3, 0, 0, 90], [0, 3, 0, -126],
[0, 0, 3, -72], [0, 0, 0, 1]]))
print('loading concept image data')
self.imgdata = masker.fit_transform(concept_images)
numpy.save(
os.path.join(self.datadir, 'imgdata_%dmm.npy' % self.resolution),
self.imgdata)
def save(self):
with open('%s/neurovault_%dmm.pkl' % (self.datadir, self.resolution),
'wb') as f:
pickle.dump(self, f)
def build_design_matrix(self, force_load=False):
if not force_load and os.path.exists(
os.path.join(self.datadir, 'desmtx.csv')):
self.desmtx = pandas.DataFrame.from_csv(
os.path.join(self.datadir, 'desmtx.csv'))
print('using cached design matrix')
return
print('building design matrix')
all_concept_pmids = []
for k in self.concept_pmids.keys():
all_concept_pmids = all_concept_pmids + self.concept_pmids[k]
all_concept_pmids = list(set(all_concept_pmids))
all_concept_pmids.sort()
all_concepts = list(self.concept_pmids.keys())
self.desmtx = pandas.DataFrame(data=0,
index=all_concept_pmids,
columns=all_concepts)
for k in self.concept_pmids.keys():
pmids = self.concept_pmids[k]
self.desmtx[k][pmids] = 1
# drop columns with too few matches
self.desmtx = self.desmtx.ix[:,
self.desmtx.sum() > self.ma_count_thresh]
self.desmtx.to_csv(os.path.join(self.datadir, 'desmtx.csv'))
""" Load a Dataset and generate a full set of meta-analysis
images--i.e., run a meta-analysis on every single feature.
"""
neurosynth_data_dir = "/home/data/nbc/misc-projects/niconn-macm/code/neurosynth/"
if not op.isfile(op.join(neurosynth_data_dir, "dataset.pkl")):
# Create Dataset instance from a database file.
dataset = Dataset(op.join(neurosynth_data_dir, "database.txt"))
# Load features from file
dataset.add_features(op.join(neurosynth_data_dir, "features.txt"))
# Pickle the Dataset to file so we can use Dataset.load() next time
# instead of having to sit through the generation process again.
dataset.save(op.join(neurosynth_data_dir, "dataset.pkl"))
# Load pickled Dataset--assumes you've previously saved it. If not,
# follow the create_a_new_dataset_and_load_features example.
dataset = Dataset.load(op.join(neurosynth_data_dir, "dataset.pkl"))
# Get the full list of feature names
feature_list = dataset.get_feature_names()
# Run a meta-analysis on each feature, and save all the results to
# a directory called results. Note that the directory will not be
# created for you, so make sure it exists.
# Here we use the default frequency threshold of 0.001 (i.e., a
# study is said to have a feature if more than 1 in every 1,000
# words is the target word), and an FDR correction level of 0.05.
out_dir = "/home/data/nbc/misc-projects/meta-gradients/code/feature_maps"
xfm2vol.inputs.identity = 'fsaverage4'
xfm2vol.inputs.hemi = 'lh'
xfm2vol.inputs.transformed_file = volume_file
xfm2vol.inputs.template_file = template
xfm2vol.run()
#make masks to input into neurosynth
def cluster2masks(clusterfile):
clustermap = nb.load(clusterfile).get_data()
for x in range(1,clustermap.max()+1):
clustermask = (clustermap==x).astype(int)
nImg = nb.Nifti1Image(clustermask, None)
nb.save(nImg, os.path.abspath(clusterfile+'_clustermask'+str(x)+'.nii'))
cluster2masks(volume_file)
dataset_file = '/home/raid3/watanabe/neurosynth/data/dataset.pkl'
if not os.path.exists(dataset_file):
dataset = Dataset('/home/raid3/watanabe/neurosynth/data/database.txt')
dataset.add_features('/home/raid3/watanabe/neurosynth/data/features.txt')
dataset.save(dataset_file)
else:
dataset = cPickle.load(open(dataset_file,'rb'))
clustermask = volume_file+'_clustermask'+str(3)+'.nii'
ids = dataset.get_ids_by_mask(clustermask)
features = dataset.feature_table.get_features_by_ids(ids)
#mri_surf2vol --identity fsaverage4 --surfval /scr/ilz1/Data/attemptsurface.nii --hemi 'lh' --o /scr/ilz1/Data/results/surf2volume.nii --template /scr/ilz1/Data/freesurfer/fsaverage4/mri/orig.mgz
