ggplot2-multivariate-scatter-visualization
Summary
Create publication-ready 2D scatter plots of multivariate data (e.g., PCA scores) colored by categorical metadata (batch, group) to visually assess sample clustering and batch correction efficacy. This skill enables side-by-side comparison of uncorrected vs. corrected metabolomics data in reduced dimensionality space.
When to use
After performing PCA (or other dimensionality reduction) on a metabolite matrix, when you need to visualize whether batch effects are present in uncorrected data, or whether a batch correction method (e.g., CordBat) successfully removes batch-driven clustering while preserving biological group structure. Trigger: you have PCA scores (PC1, PC2), sample-level batch annotations, and group annotations.
When NOT to use
- Input data has not been centered and scaled — prcomp must be run with scale=TRUE before extracting scores for visualization.
- Batch or group metadata contain missing or mismatched sample labels — all samples in PCA scores must have corresponding batch and group assignments.
- More than 2–3 principal components are needed to explain batch/group structure — scatter plot interpretation becomes unreliable; consider interactive 3D plots or other dimensionality reduction methods.
Inputs
- log2-transformed metabolite matrix (rows = samples, columns = metabolites)
- batch vector (sample-level batch labels)
- group vector (sample-level group labels)
- PCA scores data frame (PC1, PC2, batch, group)
Outputs
- ggplot2 scatter plot object colored by batch
- ggplot2 scatter plot object colored by group
- paired comparison figures (uncorrected vs. corrected)
How to apply
Extract PC1 and PC2 scores from a prcomp object and combine them with batch and group metadata into a data frame. Use ggplot2 to map PC1 to the x-axis, PC2 to the y-axis, and color aesthetic to batch or group. Generate two paired plots (uncorrected and corrected data) to visually compare how batch correction redistributes samples in PCA space. The rationale: batch-corrected data should show reduced batch-driven clustering (points colored by batch should overlap more) while retaining group separation if biological signal is preserved. Scale PCA input data (scale=TRUE in prcomp) to ensure all metabolites contribute equally.
Related tools
- prcomp (Performs PCA on scaled metabolite matrix; outputs scores (PC1, PC2, …) to be plotted)
- ggplot2 (Creates scatter plots mapping PCA scores to aesthetics (x, y, color) to visualize sample clustering by batch or group)
- CordBat (Produces corrected metabolite matrix (fit$X.cor) that is back-transformed and re-projected via PCA for comparison with uncorrected PCA plot) — https://github.com/BorchLab/CordBat
- dplyr (Data frame manipulation to combine PCA scores with metadata)
Examples
pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE); pca_df <- data.frame(PC1=pca_res$x[,'PC1'], PC2=pca_res$x[,'PC2'], batch=batch_vec, group=group_vec); ggplot(pca_df, aes(x=PC1, y=PC2, color=batch)) + geom_point(size=3) + theme_minimal()
Evaluation signals
- Uncorrected data shows tight clustering of samples by batch in PCA space (e.g., all Batch_A samples occupy one region, Batch_B another), indicating strong batch effects.
- Corrected data shows batch-driven clusters dissolved (samples from different batches intermix) while biological group separation is preserved or enhanced.
- All samples in PCA scores data frame have valid batch and group assignments with no missing values; sample counts match between PCA output and metadata.
- PC1 and PC2 axis labels report the percentage of variance explained (e.g., 'PC1 (35.2%)', 'PC2 (18.9%)'); total variance should be reasonable for metabolomics data.
- Visual inspection confirms that the corrected plot does not collapse all groups into one cluster — biological signal is retained.
Limitations
- PCA captures only linear variance structure; non-linear batch effects or confounders not aligned with principal components may not be visible.
- Visualization limited to 2D (PC1 vs. PC2); if batch/group structure exists primarily in PC3 or higher, the 2D scatter plot may be misleading.
- CordBat corrects with reference batch 'Ref' and grouping=FALSE; if grouping=TRUE, correction strategy differs and may produce different visual outcomes.
- Batch correction success (visual separation) does not guarantee statistical significance or biological validity; complementary statistical tests (e.g., PERMANOVA) are recommended.
Evidence
- [methods] Create a data frame combining PCA scores (PC1, PC2) with batch and group metadata, then plot with ggplot2 colored by batch and group to visualize uncorrected data structure.: "Create a data frame combining PCA scores (PC1, PC2) with batch and group metadata, then plot with ggplot2 colored by batch and group to visualize uncorrected data structure."
- [methods] Perform PCA on the back-transformed (2^) corrected metabolite data with scaling, combine scores with metadata, and generate ggplot2 plots colored by batch and group to visualize corrected data structure.: "Perform PCA on the back-transformed (2^) corrected metabolite data with scaling, combine scores with metadata, and generate ggplot2 plots colored by batch and group to visualize corrected data"
- [methods] pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE): "pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE)"
1---2name: ggplot2-multivariate-scatter-visualization3description: Use when after performing PCA (or other dimensionality reduction) on a metabolite matrix, when you need to visualize whether batch effects are present in uncorrected data, or whether a batch correction method (e.4license: CC-BY-4.05---67# ggplot2-multivariate-scatter-visualization89## Summary1011Create publication-ready 2D scatter plots of multivariate data (e.g., PCA scores) colored by categorical metadata (batch, group) to visually assess sample clustering and batch correction efficacy. This skill enables side-by-side comparison of uncorrected vs. corrected metabolomics data in reduced dimensionality space.1213## When to use1415After performing PCA (or other dimensionality reduction) on a metabolite matrix, when you need to visualize whether batch effects are present in uncorrected data, or whether a batch correction method (e.g., CordBat) successfully removes batch-driven clustering while preserving biological group structure. Trigger: you have PCA scores (PC1, PC2), sample-level batch annotations, and group annotations.1617## When NOT to use1819- Input data has not been centered and scaled — prcomp must be run with scale=TRUE before extracting scores for visualization.20- Batch or group metadata contain missing or mismatched sample labels — all samples in PCA scores must have corresponding batch and group assignments.21- More than 2–3 principal components are needed to explain batch/group structure — scatter plot interpretation becomes unreliable; consider interactive 3D plots or other dimensionality reduction methods.2223## Inputs2425- log2-transformed metabolite matrix (rows = samples, columns = metabolites)26- batch vector (sample-level batch labels)27- group vector (sample-level group labels)28- PCA scores data frame (PC1, PC2, batch, group)2930## Outputs3132- ggplot2 scatter plot object colored by batch33- ggplot2 scatter plot object colored by group34- paired comparison figures (uncorrected vs. corrected)3536## How to apply3738Extract PC1 and PC2 scores from a prcomp object and combine them with batch and group metadata into a data frame. Use ggplot2 to map PC1 to the x-axis, PC2 to the y-axis, and color aesthetic to batch or group. Generate two paired plots (uncorrected and corrected data) to visually compare how batch correction redistributes samples in PCA space. The rationale: batch-corrected data should show reduced batch-driven clustering (points colored by batch should overlap more) while retaining group separation if biological signal is preserved. Scale PCA input data (scale=TRUE in prcomp) to ensure all metabolites contribute equally.3940## Related tools4142- **prcomp** (Performs PCA on scaled metabolite matrix; outputs scores (PC1, PC2, …) to be plotted)43- **ggplot2** (Creates scatter plots mapping PCA scores to aesthetics (x, y, color) to visualize sample clustering by batch or group)44- **CordBat** (Produces corrected metabolite matrix (fit$X.cor) that is back-transformed and re-projected via PCA for comparison with uncorrected PCA plot) — https://github.com/BorchLab/CordBat45- **dplyr** (Data frame manipulation to combine PCA scores with metadata)4647## Examples4849```50pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE); pca_df <- data.frame(PC1=pca_res$x[,'PC1'], PC2=pca_res$x[,'PC2'], batch=batch_vec, group=group_vec); ggplot(pca_df, aes(x=PC1, y=PC2, color=batch)) + geom_point(size=3) + theme_minimal()51```5253## Evaluation signals5455- Uncorrected data shows tight clustering of samples by batch in PCA space (e.g., all Batch_A samples occupy one region, Batch_B another), indicating strong batch effects.56- Corrected data shows batch-driven clusters dissolved (samples from different batches intermix) while biological group separation is preserved or enhanced.57- All samples in PCA scores data frame have valid batch and group assignments with no missing values; sample counts match between PCA output and metadata.58- PC1 and PC2 axis labels report the percentage of variance explained (e.g., 'PC1 (35.2%)', 'PC2 (18.9%)'); total variance should be reasonable for metabolomics data.59- Visual inspection confirms that the corrected plot does not collapse all groups into one cluster — biological signal is retained.6061## Limitations6263- PCA captures only linear variance structure; non-linear batch effects or confounders not aligned with principal components may not be visible.64- Visualization limited to 2D (PC1 vs. PC2); if batch/group structure exists primarily in PC3 or higher, the 2D scatter plot may be misleading.65- CordBat corrects with reference batch 'Ref' and grouping=FALSE; if grouping=TRUE, correction strategy differs and may produce different visual outcomes.66- Batch correction success (visual separation) does not guarantee statistical significance or biological validity; complementary statistical tests (e.g., PERMANOVA) are recommended.6768## Evidence6970- [methods] Create a data frame combining PCA scores (PC1, PC2) with batch and group metadata, then plot with ggplot2 colored by batch and group to visualize uncorrected data structure.: "Create a data frame combining PCA scores (PC1, PC2) with batch and group metadata, then plot with ggplot2 colored by batch and group to visualize uncorrected data structure."71- [methods] Perform PCA on the back-transformed (2^) corrected metabolite data with scaling, combine scores with metadata, and generate ggplot2 plots colored by batch and group to visualize corrected data structure.: "Perform PCA on the back-transformed (2^) corrected metabolite data with scaling, combine scores with metadata, and generate ggplot2 plots colored by batch and group to visualize corrected data"72- [methods] pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE): "pca_res <- prcomp(cordbat_example[, metabolite_cols], scale. = TRUE)"
