This is a Pytorch implementation of Harmony algorithm on single-cell sequencing data integration. Please see Ilya Korsunsky et al., 2019 for details.

This package is published on PyPI:

pip install harmony-pytorch

Given an embedding X as a N-by-d matrix in numpy array structure (N for number of cells, d for embedding components) and cell attributes as a Data Frame df_metadata, use Harmony for data integration as the following:

from harmony import harmonize
Z = harmonize(X, df_metadata, batch_key = 'Channel')

where Channel is the attribute in df_metadata for batches.

Alternatively, if there are multiple attributes for batches, write:

Z = harmonize(X, df_metadata, batch_key = ['Lab', 'Date'])

It's easy for Harmony-pytorch to work with count matrix data structure from PegasusIO package. Let data be a MultimodalData object in Python:

from harmony import harmonize
Z = harmonize(data.obsm['X_pca'], data.obs, batch_key = 'Channel')
data.obsm['X_pca_harmony'] = Z

This will calculate the harmonized PCA matrix for the default UnimodalData of data.

Given a UnimodalData object unidata, you can also use the code above to perform Harmony algorithm: simply substitute unidata for data there.

It's easy for Harmony-pytorch to work with annotated count matrix data structure from anndata package. Let adata be an AnnData object in Python:

from harmony import harmonize
Z = harmonize(adata.obsm['X_pca'], adata.obs, batch_key = '<your-batch-key>')
adata.obsm['X_harmony'] = Z

where <your-batch-key> should be replaced by the actual batch key attribute name in your data.

For details about AnnData data structure, please refer to its documentation.