Batch Effects & Ordination
Interactive UMAP scatter of sample composition — toggle coloring by body site, data source, or sequencing modality; compare uncorrected vs. batch-corrected projections. Coordinates shown are synthetic illustrations; replace with real PCoA/UMAP computed from the genus relab tables.
These are synthetic coordinates for UI demonstration. To generate real embeddings: compute Bray–Curtis PCoA or UMAP from corpus_16s_genus_relab_expanded.parquet, optionally after ComBat-seq or MMUPHin batch correction keyed on source.
Color by:
Coordinates:
Interpretation Guide
What to look for: Body Site

In a well-structured ordination, stool/gut samples should form a tight cluster distinct from oral, skin, and urogenital. Overlap between stool and oral may indicate taxonomic cross-contamination or a mouth-gut axis signal.

What to look for: Data Source

Before correction — clusters often separate by source (MGnify_full, cMD, AGP) due to sequencing platform differences, amplicon regions (V4 vs other), and bioinformatic pipelines.
After correction — source-driven separation should decrease while body-site clustering is preserved.

Batch Correction Methods

Recommended approaches for this corpus: MMUPHin (meta-analysis MaAsLin2 framework), ComBat-seq (count-based negative-binomial), or ConQuR (conditional quantile regression). Key batch variable: source; optional co-variate: sequencing_platform.