Linear OT

Linear OT implementation for domain adaptation for aligning parallel unpaired scRNA/scATAC

Dependencies

- numpy
- scipy

Usage

Inputs are

1. xs, a matrix of "source" RNA expression levels from scRNA-seq
2. xt, a matrix of "target" gene activities calcualated from ATAC-seq peak accessibility

Also, optionally:

1. ws: vector of weights representing the size of each cluster/metacell from RNA
2. wt: same as above but for ATAC
3. rho: float in range [0,1] representing whether the final transformation should be closer to RNA distribution (0) or ATAC distribution (1). Default value is 1.
4. reg: small float to make sure covariance matrices are invertible

xs and xt should be dense matrices representing expression of highly variable genes.

To get transformed aligned xs and xt, run the following:

from barycenter import LinearOT

model = LinearOT()

xs_transformed, xt_transformed = model.fit_transform(xs, xt)