This is the maintained and official version of neuroCombat (previously hosted here) introduced in our our recent paper.
neuroCombat is hosted on PyPI, and the easiest way to install neuroCombat is to use the pip
command:
pip install neuroCombat
The neuroCombat
function performs harmonization
from neuroCombat import neuroCombat
import pandas as pd
import numpy as np
# Getting example data
# 200 rows (features) and 10 columns (scans)
data = np.genfromtxt('testdata/testdata.csv', delimiter=",", skip_header=1)
# Specifying the batch (scanner variable) as well as a biological covariate to preserve:
covars = {'batch':[1,1,1,1,1,2,2,2,2,2],
'gender':[1,2,1,2,1,2,1,2,1,2]}
covars = pd.DataFrame(covars)
# To specify names of the variables that are categorical:
categorical_cols = ['gender']
# To specify the name of the variable that encodes for the scanner/batch covariate:
batch_col = 'batch'
#Harmonization step:
data_combat = neuroCombat(dat=data,
covars=covars,
batch_col=batch_col,
categorical_cols=categorical_cols)["data"]
-
eb
:True
orFalse
. Should Empirical Bayes be performed? IfFalse
, the harmonization model will be fit for each feature separately. This is equivalent to performing a location/shift (L/S) correction to each feature separately (no information pooling across features). -
parametric
:True
orFalse
. Should parametric adjustements be performed?True
by default. -
mean_only
:True
orFalse
. Should only be means adjusted (no scaling)?False
by default -
ref_batch
: batch name to be used as the reference batch for harmonization.None
by default, in which case the average across scans/images/sites is taken as the reference batch.
Since version 0.2.10, the neuroCombat
function outputs a dictionary with 3 elements:
data
: A numpy array of the harmonized data, with the same dimension (shape) as the input data.estimates
: A dictionary of the neuroCombat estimates; useful for visualization and understand scanner effects.info
: A dictionary of the inputs needed for ComBat harmonization (batch/scanner information, etc.)
To simply return the harmonized data, one can use the following:
data_combat = neuroCombat(dat=dat, ...)["data"]
where ...
are the user-specified arguments needed for harmonization.