dm.heatmap() ######################################################################### # Load and Z-score a Covariates File # ---------------------------------- # # Now we're going to handle a covariates file that's been generated by a preprocessing routine. # First we'll read in the text file using pandas and convert it to a design matrix. # To be explicit with the meta-data we're going to change some default attributes during conversion. import pandas as pd covariatesFile = os.path.join(get_resource_path(),'covariates_example.csv') cov = pd.read_csv(covariatesFile) cov = Design_Matrix(cov,hasIntercept=False,hrf=[]) cov.heatmap() ######################################################################### # The class has several methods features for basic data scaling and manipulation. Others can likely be found in pandas core functionality. # Here we fill NaN values with 0 and zscore all columns except the last. Because the class has all of pandas functionality, method-chaining is built-in. cov = cov.fillna(0).zscore(cov.columns[:-1]) cov.heatmap() ######################################################################### # Concatenate Multiple Design Matrices # ---------------------------------- # # A really nice feature of this class is simplified, but intelligent matrix concatentation. Here it's trivially to horizontally concatenate our convolved onsets and covariates, while keeping our column names and order. full = dm.append(cov,axis=1)
sampling_freq=1. / TR,
columns=['face_A', 'face_B', 'house_A', 'house_B'])
#########################################################################
# Notice how this look exactly like a pandas dataframe. That's because design matrices are *subclasses* of dataframes with some extra attributes and methods.
print(dm)
#########################################################################
# Let's take a look at some of that meta-data. We can see that no columns have been convolved as of yet and this design matrix has no polynomial terms (e.g. such as an intercept or linear trend).
print(dm.details())
#########################################################################
# We can also easily visualize the design matrix using an SPM/AFNI/FSL style heatmap
dm.heatmap()
#########################################################################
# Adding nuisiance covariates
# ---------------------------
#
# Legendre Polynomials
# ********************
#
# A common operation is adding an intercept and polynomial trend terms (e.g. linear and quadtratic) as nuisance regressors. This is easy to do. Consistent with other software packages, these are orthogonal Legendre poylnomials on the scale -1 to 1.
# with include_lower = True (default), 2 here means: 0-intercept, 1-linear-trend, 2-quadtratic-trend
dm_with_nuissance = dm.add_poly(2, include_lower=True)
dm_with_nuissance.heatmap()
#########################################################################
onsets = pd.read_csv(layout.get(subject=subject, suffix='events')[0].path, sep='\t')
onsets.columns = ['Onset', 'Duration', 'Stim']
return onsets_to_dm(onsets, sampling_freq=1/tr, run_length=n_tr)
dm = load_bids_events(layout, 'S01')
motor_variables = ['video_left_hand','audio_left_hand', 'video_right_hand', 'audio_right_hand']
ppi_dm = dm.drop(motor_variables, axis=1)
ppi_dm['motor'] = pd.Series(dm.loc[:, motor_variables].sum(axis=1))
ppi_dm_conv = ppi_dm.convolve()
ppi_dm_conv['vmpfc'] = vmpfc
ppi_dm_conv['vmpfc_motor'] = ppi_dm_conv['vmpfc']*ppi_dm_conv['motor_c0']
dm = Design_Matrix(pd.concat([ppi_dm_conv, csf, mc_cov, spikes.drop(labels='TR', axis=1)], axis=1), sampling_freq=1/tr)
dm = dm.add_poly(order=2, include_lower=True)
dm.heatmap()
Okay, now we are ready to run the regression analysis and inspect the interaction term to find regions where the connectivity profile changes as a function of the motor task.
We will run the regression and smooth all of the images, and then examine the beta image for the PPI interaction term.
smoothed.X = dm
ppi_stats = smoothed.regress()
vmpfc_motor_ppi = ppi_stats['beta'][int(np.where(smoothed.X.columns=='vmpfc_motor')[0][0])]
vmpfc_motor_ppi.plot()
This analysis tells us which regions are more functionally connected with the vmPFC during the motor conditions relative to the rest of experiment.
We can make a thresholded interactive plot to interrogate these results, but it looks like it identifies the ACC/pre-SMA.
columns=['stim_A','stim_B','stim_C','stim_D']
)
#########################################################################
# Notice how this look exactly like a pandas dataframe. That's because design matrices are *subclasses* of dataframes with some extra attributes and methods.
print(dm)
#########################################################################
# Let's take a look at some of that meta-data. We can see that no columns have been convolved as of yet and this design matrix has no polynomial terms (e.g. such as an intercept or linear trend).
print(dm.details())
#########################################################################
# We can also easily visualize the design matrix using an SPM/AFNI/FSL style heatmap
dm.heatmap()
#########################################################################
# A common operation might include adding an intercept and polynomial trend terms (e.g. linear and quadtratic) as nuisance regressors. This is easy to do. Note that polynomial terms are normalized to unit variance (i.e. mean = 0, std = 1) before inclusion to keep values on approximately the same scale.
# with include_lower = True (default), 1 here means: 0-intercept, 1-linear-trend, 2-quadtratic-trend
dm_with_nuissance = dm.add_poly(2,include_lower=True)
dm_with_nuissance.heatmap()
#########################################################################
# We can see that 3 new columns were added to the design matrix. We can also inspect the change to the meta-data. Notice that the Design Matrix is aware of the existence of three polynomial terms now.
print(dm_with_nuissance.details())
#########################################################################
# Polynomial variables are not the only type of nuisance covariates that can be generate for you. Design Matrix also supports the creation of discrete-cosine basis functions ala SPM. This will create a series of filters added as new columns based on a specified duration, defaulting to 180s.
