pcaMethods

Dependencies & Environment

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

• Version: 2.4.0 · Bioconductor: 3.23 · R: ≥ 4.6
• Depends: Biobase
• Imports: BiocGenerics, Rcpp, MASS
• System requirements: Rcpp
• Install: BiocManager::install("pcaMethods")

When to Use

• Handling Outliers: When performing PCA on metabolite or microarray data corrupted with extreme values using a robust singular value decomposition (method="robustPca").
• Missing Value Imputation: When estimating missing values (NA) in incomplete datasets using Probabilistic PCA (method="ppca") or Nipals (method="nipals").
• Algorithm Comparison: When comparing multiple PCA formulations (e.g., "svd", "ppca", "bpca", "svdImpute", "nipals", "nlpca") on the same ExpressionSet or matrix using a unified pca() interface.

When NOT to Use

• Sparse Single-Cell Data: For single-cell RNA-seq data with high dropout rates, use scry or GLM-PCA instead because they model the count distribution directly.
• Complete Datasets: For standard, complete matrices without missing values or outliers, use base R's prcomp or irlba because they are faster and computationally lighter.

Data Requirements

• Input Format: A numeric matrix, data.frame, or an ExpressionSet (from the Biobase package).
• Missing Values: Missing data should be represented as NA.
• Normalization: Data is typically mean-centered and scaled (e.g., using the prep() function with scale="none" and center=TRUE) prior to running PCA.

Key Parameters

• method ("svd"): The PCA algorithm to use (e.g., "svd", "robustPca", "ppca", "bpca", "svdImpute", "nipals", "nlpca").
• nPcs (2): Number of principal components to calculate.
• center (TRUE): Logical indicating whether to mean-center the variables.
• scale ("none"): Scaling method to apply in the prep() function.
• maxSteps (1000): Maximum number of iterations for iterative algorithms like NLPCA.
• fold (10): Number of folds for cross-validation when using the Q2() function.
• evalPcs (1:5): The principal components to evaluate when estimating error with kEstimate().
• em ("nrmsep"): The error metric to use during cross-validation in kEstimate().

Best Practices

• Pre-processing: Always pre-process data using the prep() function to handle centering and scaling before running pca().
• Determine Components: Use cross-validation via the Q2() function to evaluate the optimal number of principal components to retain.
• Outlier Management: When dealing with extreme outliers, set them to NA and use method="ppca", or use method="robustPca" directly on the complete data.
• Error Estimation: Estimate imputation error using kEstimate() with cross-validation to ensure the chosen method and number of PCs are appropriate.

Common Pitfalls

• Failing to Center Before Outlier Injection: Creating artificial outliers for testing on un-centered data, which shifts original means and prevents objective comparison. Fix: Use scale(..., center=TRUE, scale=FALSE) before injecting outliers.
• PPCA Instability: PPCA failing to converge or misestimating values in rare cases. Fix: Use a while loop checking if sum(abs(Q2)) > 1 to re-run the Q2() function if PPCA becomes unstable.
• Non-Numeric Input: Passing a matrix containing character columns or factors to pca(), causing the algorithms to fail. Fix: Ensure the input matrix or ExpressionSet contains only numeric values.

Alternatives

• missMDA: For PCA and imputation on incomplete datasets using regularized iterative PCA.
• impute: For K-Nearest Neighbors imputation (impute.knn) of missing microarray data.
• FactoMineR: For advanced exploratory multivariate data analysis on complete datasets.
• irlba: For fast, truncated SVD and PCA on extremely large, sparse matrices.

Citations

• Liu, L., Hawkins, D.M., Ghosh, S. and Young, S.S. Robust singular value decomposition analysis of microarray data. PNAS, 2003;100:13167–13172.
• Hawkins, D.M., Liu, L. and Young, S.S. Robust Singular Value Decomposition. National Institute of Statistical Sciences, 2001, Tech Report 122.

References

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

<!-- biomate-cta -->

---

Run this on BioMate

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

▶ Open pcamethods on BioMate →