| name | bioconductor-sva |
| description | The sva package contains functions for removing batch effects and other unwanted variation in high-throughput experiment. Specifically, the sva package contains functions for the identifying and building surrogate variables for high-dimensi |
| when_to_use | Use when: Surrogate Variable Analysis (SVA): Identifying and adjusting for hidden, unmodeled, or latent sources of variation in high-dimensional datasets using sva and num.sv.; Known Batch Effect Correction (ComBat): Adjusting for known batch effects in normalized gene expression data using ComBat.; RNA-seq Count Correction (ComBat-seq): Removing batch effects from raw RNA-seq count data using ComBat_seq wh. Not for: For single-cell RNA-seq batch correction, use Seurat or batchelor instead because ComBat/sva can overcorrect and destroy single-cell biological heterogeneity.; When the batch effect is completely confounded with the biological variable of interest, d |
| user-invocable | false |
sva
Dependencies & Environment
Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.
- Version: 3.60.0 · Bioconductor: 3.23 · R: ≥ 4.6
- Depends: mgcv, genefilter, BiocParallel
- Imports: matrixStats, limma, edgeR
- System requirements: URL
- Install:
BiocManager::install("sva")
When to Use
- Surrogate Variable Analysis (SVA): Identifying and adjusting for hidden, unmodeled, or latent sources of variation in high-dimensional datasets using
sva and num.sv.
- Known Batch Effect Correction (ComBat): Adjusting for known batch effects in normalized gene expression data using
ComBat.
- RNA-seq Count Correction (ComBat-seq): Removing batch effects from raw RNA-seq count data using
ComBat_seq while preserving the integer count structure.
- Sequencing Data SVA: Estimating surrogate variables specifically for sequencing data using
svaseq.
- Frozen SVA for Prediction: Removing batch effects for prediction and clustering on new test datasets using
fsva.
When NOT to Use
- For single-cell RNA-seq batch correction, use
Seurat or batchelor instead because ComBat/sva can overcorrect and destroy single-cell biological heterogeneity.
- When the batch effect is completely confounded with the biological variable of interest, do not use ComBat/sva because it will remove the biological signal.
Data Requirements
- Expression Data: A matrix of normalized gene expression values (e.g., from
exprs) or raw integer counts (specifically for ComBat_seq).
- Design Matrix: A model matrix representing the biological variables of interest (full model) created with
model.matrix.
- Batch Covariate: A vector or factor indicating the known batch for each sample.
Key Parameters
- dat: The gene expression matrix (genes in rows, samples in columns).
- mod: Model matrix of interest representing the biological variables to preserve.
- mod0 (NULL): Null model matrix containing only the adjustment variables.
- n.sv (NULL): Number of surrogate variables to estimate; can be estimated using
num.sv.
- batch: Factor or vector specifying the known batch variable.
- covar_mod (NULL): Model matrix of covariates to adjust for during
ComBat_seq run.
- par.prior (TRUE): Whether to use parametric empirical Bayesian adjustments in
ComBat.
- group (NULL): Biological variable to preserve in
ComBat_seq.
Best Practices
- Always specify the biological model of interest (
mod) when running sva or ComBat to prevent the algorithm from accidentally removing the biological signal.
- For downstream differential expression with
limma, include the estimated surrogate variables as covariates in the design formula (e.g., using lmFit and contrasts.fit).
- Use
ComBat_seq instead of standard ComBat when working with raw RNA-seq counts to maintain the integer distribution required by count-based differential tools.
- Use
num.sv to automatically estimate the number of latent factors before running sva.
Common Pitfalls
- ComBat on raw counts: Running standard
ComBat on raw counts; fix this by using ComBat_seq for raw counts.
- Missing biological model: Failing to include the biological design matrix in
ComBat, which leads to over-correction; fix this by always passing the mod parameter.
- Confounded designs: Attempting to correct batch effects when they are completely confounded with the biological group; fix this by avoiding ComBat/SVA entirely.
Alternatives
limma: Provides removeBatchEffect for linear model-based batch correction, ideal for visualization but not for downstream differential testing.ruv (Remove Unwanted Variation): Uses control genes or replicate samples to estimate and remove unwanted variation.harman: A PCA-based batch correction method that constrains the correction to prevent over-correction of biological variation.
Citations
- Leek, J. T., & Storey, J. D. (2007). Capturing heterogeneity in gene expression studies by surrogate variable analysis. PLoS Genetics, 3(9), e161.
- Johnson, W. E., Li, C., & Rabinovic, A. (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics, 8(1), 118-127.
- Zhang, Y., Parmigiani, G., & Johnson, W. E. (2020). ComBat-seq: batch effect adjustment in RNA-seq count data using empirical Bayes methods. NAR Genomics and Bioinformatics, 2(3), lqaa078.
References
Run this on BioMate
This skill is the knowledge layer — when, why, and how to use sva. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.
▶ Open sva on BioMate →
