Figure 1

Unsupervised analysis of the top 250 methylation variable positions (MVPs) in formalin-fixed, paraffin wax-embedded (FFPE) human papillomavirus-positive (HPV+) and HPV-negative (HPV-) tumor samples. (A) Singular value decomposition: PC-k denotes the kth principal component, DA denotes survival at censoring date. The first two principal components PC-1 and PC-2 most strongly correlated with HPV status, whereas the remaining significant components associated with clinical parameters, including alcohol consumption, smoking, age, sex, tumor stage, and grade. No association was found with technical factors (such as Sentrix position and Sentrix ID) on the array. (B) The first two principal components clearly distinguish HPV status. HPV+ samples are plotted in black, HPV- samples are in red, and m/f indicates male/female. (C) Clusters inferred by the unsupervised consensus-clustering algorithm for the top 250 MVPs as found using the MAD estimator.