今日云讲堂 | 李婧翌：ClusterDE a post-clustering multiple-testing ……

In standard single-cell gene-expression data analysis, cells are annotated as candidate cell types in two steps: given a cell-by-gene matrix, cells are first clustered, and then each cluster is assigned with marker genes that have higher expression levels in the cluster than other clusters. Hence, the procedure is clustering followed by multiple testing on the same data, a phenomenon referred to as ``double dipping." Due to double dipping, every cell cluster would inevitably have marker genes assigned even if the cluster should not be separated from the other cells, resulting in false positive cluster-marker genes. To overcome this challenge, we propose ClusterDE, a post-clustering multiple-testing framework, for controlling the false discovery rate (FDR) regardless of clustering quality. The core idea of ClusterDE is to set up in silico negative control data. Applied to single-cell gene expression data, we show that ClusterDE not only has solid FDR control but also finds cell-type marker genes that are biologically meaningful. ClusterDE is fast, transparent, and adaptive to a wide range of clustering methods and statistical tests.