cyCombinePy

Python port of cyCombine for batch correction of single-cell cytometry data.

cyCombinePy is AnnData-native and reuses existing Python libraries instead of reimplementing primitives:

• ComBat: inmoose.pycombat
• SOM clustering: FlowSOM
• FCS I/O: pytometry / readfcs
• Batch-effect metrics: scib-metrics

Pipeline

The cyCombine workflow ports over unchanged:

1. Batch-wise normalize expression per marker (cycombinepy.normalize)
2. Self-organizing map clustering of cells (cycombinepy.create_som)
3. Per-cluster ComBat correction with optional covariates and anchors (cycombinepy.correct_data)

Step 1 operates on a normalized view so clusters represent biology rather than batch. Step 3 is applied to the unnormalized data per cluster so rare populations aren't over-corrected.

Quickstart

import cycombinepy as pc

# 1. Load FCS files into AnnData
adata = pc.io.read_fcs_dir(
    "data/",
    metadata="metadata.csv",
    batch_key="Batch",
    sample_key="Patient",
    condition_key="condition",
    cofactor=5,           # asinh cofactor for CyTOF
)

# 2. Inspect batch effects before correction
figs = pc.detect_batch_effect_express(adata, out_dir="before/")

# 3. End-to-end batch correction
pc.batch_correct(
    adata,
    xdim=8, ydim=8,
    covar="condition",
)
# Corrected matrix is now in adata.layers["cycombine_corrected"]

# 4. Evaluate
from cycombinepy.correct import CORRECTED_LAYER
uncorr = pc.compute_emd(adata, cell_key="cycombine_som")
corr   = pc.compute_emd(adata, cell_key="cycombine_som", layer=CORRECTED_LAYER)
report = pc.evaluate_emd(uncorr, corr)
print(report.groupby("marker")["reduction_pct"].mean())

Or use the modular API:

pc.transform_asinh(adata, cofactor=5)
pc.normalize(adata, method="scale")
pc.create_som(adata, xdim=8, ydim=8)
pc.correct_data(adata, label_key="cycombine_som", covar="condition")

Public API

FCS I/O lives in cycombinepy.io, plotting in cycombinepy.plotting, and an optional scib_metrics wrapper in cycombinepy.evaluate.

Installation

pip install -e ".[all,dev]"

Data structure conventions

• adata.X: cells × markers expression (post-asinh, pre-correction)
• adata.obs["batch"]: batch assignment (required)
• adata.obs["sample"], adata.obs["condition"], adata.obs["anchor"]: optional metadata
• adata.obs["cycombine_som"]: SOM cluster labels (written by create_som)
• adata.layers["cycombine_corrected"]: corrected expression (written by correct_data / batch_correct)

Citation

If you use cyCombinePy please cite the original cyCombine paper:

Pedersen, C.B., Dam, S.H., Barnkob, M.B., et al. cyCombine allows for robust integration of single-cell cytometry datasets within and across technologies. Nat Commun 13, 1698 (2022). https://doi.org/10.1038/s41467-022-29383-5