vsn

Dependencies & Environment

Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.

• Version: 3.80.0 · Bioconductor: 3.23 · R: ≥ 4.6
• Depends: Biobase
• Imports: affy, limma, lattice, ggplot2
• Install: BiocManager::install("vsn")

When to Use

• Microarray Normalization: Normalizing and variance-stabilizing unnormalised microarray intensity data (e.g., two-colour spotted cDNA arrays or single-colour arrays) using justvsn or vsn2.
• Affymetrix Preprocessing: Applying variance stabilization, background correction, and transformation to Affymetrix genechip data via the vsnrma wrapper.
• Reference-Based Normalization: Normalizing new test arrays against an existing reference dataset without altering the reference model using the reference argument in vsn2.
• Spike-in Calibration: Fitting calibration and transformation parameters on a subset of features (like spike-in probes) using vsn2, and then applying the model to the complete dataset using predict.

When NOT to Use

• For summarizing Affymetrix data without VSN's specific variance stabilization, use rma from the affy package because it provides standard robust multi-array average summarization.
• For addressing variance dependencies on factors other than the mean intensity (e.g., gene-inherent properties or sample-inherent transcriptional control tightness), use specialized models because VSN only addresses the dependence of variance on the mean intensity.
• For reading and processing raw image quantitation files directly into R, use read.maimages from the limma package because vsn expects already-imported data structures.

Data Requirements

• Input Format: An ExpressionSet, AffyBatch (from the affy package), RGList (from the limma package), NChannelSet, or a raw numeric matrix.
• Data Structure: Rows represent features (e.g., spots, probes) and columns represent samples or arrays.
• Normalization State: Unnormalised, raw intensity data (do not log-transform prior to VSN, as it applies its own generalized logarithm, $glog_2$, transformation).

Key Parameters

• strata: Allows choosing different offset and scaling factors for different groups of rows (e.g., print-tip groups or sectors).
• calib: Controls the calibration behavior (e.g., choosing the same offset and scaling factor throughout if calibration was already done).
• backgroundsubtract (FALSE): When set to TRUE in justvsn for an RGList, subtracts local background estimates from the incoming data.
• lts.quantile: Controls the robustness of the parameter estimation algorithm; setting it to 1 corresponds to least sum of squares regression (no outlier removal), useful for trusted spike-in sets.
• reference: Allows passing an existing fitted vsn object to normalize new arrays against a reference dataset.
• ranks (TRUE): Used in meanSdPlot to distribute data evenly along the x-axis by rank rather than average intensity.

Best Practices

• Verify Variance Stabilization: Use meanSdPlot after normalization to verify variance stabilization by checking if the running median of the standard deviation is approximately a horizontal line.
• Two-Step Fitting: When fitting parameters on a subset of data (e.g., spike-ins) using vsn2, apply the model to the complete dataset using the predict method.
• Background Correction: Avoid subtracting local background estimates unless there is actual local variability (like a spatial gradient), as VSN already estimates and subtracts an overall background estimate.
• Affymetrix Workflows: For Affymetrix genechip data, use the vsnrma wrapper to seamlessly combine VSN's background correction, between-array normalization, and transformation with RMA summarization.

Common Pitfalls

• Optimization Convergence Failure: Encountering the error "L-BFGS-B needs finite values of 'fn'" due to unsuitable data or flat parameter space directions. Fix: Check data quality, experimental design, or prior preprocessing steps for incompatible measurements.
• Systematic Trends Post-Normalization: Observing an overall trend in the meanSdPlot after normalization. Fix: Investigate raw data integrity or inadequate prior preprocessing, as this indicates the variance stabilization failed.
• Increased Random Noise: Adding random noise to the signal by unnecessarily subtracting local background estimates. Fix: Rely on VSN's overall background estimate and leave backgroundsubtract=FALSE unless spatial gradients are explicitly present.
• Metadata Loss During Coercion: Losing sample metadata when justvsn converts an RGList to an NChannelSet. Fix: Manually construct and assign an AnnotatedDataFrame to the phenoData slot of the resulting object.

Alternatives

• affy: Provides the standard rma function for Affymetrix genechip preprocessing without VSN's specific generalized logarithm transformation.
• limma: Provides read.maimages for importing data and lmFit/eBayes for linear modeling and differential expression, often used downstream of VSN.
• quantreg: Provides quantile regression, which is a more rigorous alternative to the simple running median used in meanSdPlot for assessing variance.

Citations

• Huber W, et al. (2002). Variance stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics, 18(Suppl 1), S96-S104.
• Rocke DM, Durbin B (2001). A model for measurement error for gene expression arrays. Journal of Computational Biology, 8(6), 557-569.
• Irizarry RA, et al. (2003). Summaries of Affymetrix Microarray Probe Level Data. Nucleic Acids Research, 31(4), e15.

References

• Homepage: https://bioconductor.org/packages/vsn
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/vsn/inst/doc/vsn.pdf

<!-- biomate-cta -->

---

Run this on BioMate

This skill is the knowledge layer — when, why, and how to use vsn. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.

▶ Open vsn on BioMate →