Selective Inference for Hierarchical Clustering

Classical tests for a difference in means control the Type I error rate when groups are defined priori. However, instead via clustering, then applying classical test yields an extremely inflated rate. Notably, this problem persists even if two separate and independent datasets used to define their means. To address problem, article, we propose selective inference approach between clusters. Our procedure controls by accounting fact that choice of null hypothesis was made based on data. We describe how efficiently compute exact p-values clusters obtained using agglomerative hierarchical clustering with many commonly linkages. apply our method simulated data single-cell RNA-sequencing Supplementary materials article available online.