Summary
This technical report discusses the handling of data containing outliers, particularly in the context of Principal Component Analysis (PCA). It outlines the challenges posed by extreme values in microarray or metabolite data, which can significantly affect the results of standard Singular Value Decomposition (SVD). The report introduces robustSvd, a method designed to perform SVD that is resistant to outliers, and robustPca, an implementation that utilizes robustSvd. The document details the methodology for addressing outliers, including the suggestion to remove obvious outliers before estimating PCA solutions. It also presents an example demonstrating the impact of outliers on PCA results and compares standard SVD with robust SVD and other methods. The report references previous work on robust singular value decomposition and provides a script contributed by Kevin Wright for practical application. Overall, the report serves as a guide for researchers dealing with outlier data in PCA.
