Intelligence

Analysis

cyCombine: Correct Batch Effects Without Erasing the Biological Question

The historical cyCombine study separates technical alignment from biological preservation. A cleaner combined plot still needs a defensible study design.

Technical Review of a peer-reviewed original study in Nature Communications (2022-03-31). Historical or study-specific evidence; original CFCM educational appraisal.

Start with the measurement question

A dataset can look better aligned after correction while becoming harder to interpret biologically. The first question is what variation the experiment was designed to preserve.

What the primary paper contributes

Pedersen and colleagues describe cyCombine, using normalized expression to co-cluster cells and applying ComBat within self-organizing-map groups. Its panel-merging component imputes markers absent from one panel. The authors distinguish reduction of technical variation from conservation of biological variation using separate metrics. They also identify a central limitation: batch and experimental condition cannot be completely confounded. These are historical research demonstrations, not validation of a clinical assay or a current software release. Read the primary article (DOI 10.1038/s41467-022-29383-5).

Original CFCM conceptual checklist: Design; Correct; Verify biology. No patient data or source figure reproduced.
Original CFCM conceptual diagram. Educational synthesis; not clinical validation.

A proposed local review

Before correction, write down the biological comparison and the variables that encode processing batches. Retain an uncorrected reference and define which populations or relationships must survive. Treat an attractive embedding as a visualization, not an acceptance criterion.

Keep the limits visible

For a local review, examine uncommon populations and differences that plausibly reflect biology. If disease status coincides perfectly with batch, a numerical correction cannot by itself identify which component caused the difference. Describe that design limitation rather than hiding it behind a harmonized plot.

Daily Learning Pill

Alignment and biological preservation are separate questions. Review both before interpreting a combined dataset.

Explore CFCM Resources and Troubleshooting. Review by ChatGPT/Codex, not independent scientific review or clinical validation.