Batch Effect Visualization

batch-effect-visualization

Summary

Visualize batch effects and signal drift in metabolomics data before and after normalisation using PCA and run-order plots. This skill enables rapid assessment of whether hierarchical RUV normalisation has successfully removed unwanted variation across experimental batches.

When to use

Apply this skill when you have multi-batch metabolomics data in SummarizedExperiment format and need to demonstrate that batch clustering or run-order signal drift persists in raw/imputed assays but is eliminated after hRUV normalisation (intra-batch loess + RUV-III, followed by inter-batch concatenation).

When NOT to use

• Input is already batch-corrected or is single-batch data; batch effect may be imperceptible or trivial.
• rawImpute or loessShort_concatenate assays have not been generated following the prescribed hRUV workflow (log transformation, cleaning, loess+RUV-III, concatenation); comparison will be invalid.
• colData lacks batch_info or equivalent batch labelling; PCA colouring and interpretation becomes impossible.

Inputs

• SummarizedExperiment object with rawImpute assay (log2-transformed, cleaned with threshold=0.5, intersect method)
• SummarizedExperiment object with loessShort_concatenate assay (intra-batch loess + RUV-III k=5 normalised, inter-batch concatenated)
• colData containing batch_info factor for grouping

Outputs

• PCA plot of rawImpute assay coloured by batch_info (showing batch effect)
• PCA plot of loessShort_concatenate assay coloured by batch_info (showing batch removal)
• Run-order plot (hRUV::plotRun) for selected metabolite from rawImpute (showing signal drift)
• Run-order plot (hRUV::plotRun) for same metabolite from loessShort_concatenate (showing drift correction)

How to apply

Generate PCA plots using hRUV::plotPCA on both the rawImpute assay (pre-normalisation) and loessShort_concatenate assay (post-normalisation), colouring by batch_info to reveal batch clustering. Simultaneously generate hRUV::plotRun diagnostic plots for individual metabolites (e.g., 1-methylhistamine, GlucosePos2) on both assays to reveal signal drift across run order. Visual comparison of before/after plots confirms success: rawImpute shows strong batch separation and metabolite run-order variation; loessShort_concatenate shows dispersed, batch-agnostic clustering and flat run-order profiles. Rationale: PCA reveals global batch structure; run plots reveal local metabolite-level drift within batch sequences, together providing evidence that both inter- and intra-batch unwanted variation has been removed.

Related tools

• hRUV (Provides plotPCA and plotRun diagnostic plotting functions; used to generate before/after batch-effect visualisations) — https://github.com/SydneyBioX/hRUV
• SummarizedExperiment (Container for multi-assay metabolomics data (rawImpute, loessShort_concatenate) with sample metadata; enables assay subsetting and colData extraction for plotting)
• R (Execution environment for hRUV plotting functions and visual inspection)
• dplyr (Data manipulation for subsetting metabolites and samples prior to plotting)

Examples

library(hRUV); plotPCA(dat_list[[1]], assay='rawImpute', colour_by='batch_info'); plotPCA(dat_list[[1]], assay='loessShort_concatenate', colour_by='batch_info'); plotRun(dat_list[[1]], assay='rawImpute', metabolite='1-methylhistamine'); plotRun(dat_list[[1]], assay='loessShort_concatenate', metabolite='1-methylhistamine')

Evaluation signals

• PCA plot of rawImpute shows tight clustering of samples by batch_info (strong batch effect visible as distinct colour clusters); PCA of loessShort_concatenate shows dispersed, colour-mixed clustering (batch effect removed).
• Run-order plots for rawImpute display systematic increase or decrease in metabolite signal intensity across run order (signal drift); loessShort_concatenate run plots show flat, scattered profiles (drift corrected).
• Visual inspection confirms that batch-related principal components in rawImpute (typically PC1 or PC2) account for large variance; in loessShort_concatenate, batch no longer segregates samples on principal axes.
• Both before and after plots use identical metabolite selection, y-axis scaling, and batch colour scheme to enable fair visual comparison.
• No assay contains missing values or NA entries in the plotted metabolite/sample intersection; plotting functions execute without error.

Limitations

• PCA visualisation is 2D projection; batch structure in higher-dimensional space may not be fully captured. Run plots reveal only single-metabolite drift and do not reveal multivariate batch interactions.
• Visual assessment is subjective; quantitative batch-effect metrics (e.g., silhouette width, inertia ratio, RLE) are not produced by this skill alone.
• Effectiveness depends critically on correct application of preceding hRUV workflow steps (log transformation, cleaning with threshold=0.5 and intersect method, loess smoothing, RUV-III k=5, hierarchical concatenation). Deviations will produce invalid or misleading plots.
• Requires multi-batch structure with embedded intra-batch and inter-batch replicates; single-batch or poorly replicated designs may not show batch effects warranting visualisation.
• Metabolite selection for run plots is manual; choice of 'representative' metabolites showing strong drift (e.g., 1-methylhistamine, GlucosePos2) may bias visual conclusions if drift is heterogeneous across metabolites.

Evidence

• [intro] PCA visualisation shows batch clustering before and after normalisation: "PCA visualisation of rawImpute assay shows strong batch effect, whereas PCA of the loessShort_concatenate normalised assay no longer displays batch effect when coloured by batch_info."
• [intro] Generate hRUV::plotPCA with batch colouring: "Generate a PCA plot using hRUV::plotPCA with rawImpute assay colored by batch_info to visualize the pre-normalization batch effect."
• [intro] Run plots reveal metabolite-level signal drift: "Generate hRUV::plotRun diagnostic plots for the rawImpute assay showing run-order variation for 1-methylhistamine and GlucosePos2."
• [intro] Drift correction confirmed by comparing before/after run plots: "Generate hRUV::plotRun diagnostic plots for the loessShort_concatenate assay showing the same metabolites to confirm drift and batch effects are corrected."
• [readme] hRUV hierarchical normalisation with RUV-III: "hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy with use of samples replicates in large-scale studies. The tool utilises 2 types of replicates:"