Peer Review Report For: Finding stable clusterings of single-cell RNA-seq data [version 1; peer review: 1 approved with reservations]

Résumé fourni par la source

Background There is evidently no consensus on how to find stable clusterings of cells in scRNA-seq UMI count matrices. Methods Run a count matrix through a pipeline to obtain n cell clusters. Suppose that counts for more cells from the same experiment become available. Would including them change the result? Form the matrix containing both sets of counts, obtain n clusters, restrict this clustering to the initial cells and compare it with the initial clustering. If they are not consistent, conclude that the initial clustering is unstable. This is unrealistic, but reverse the perspective: given a clustering, process samples of half of the cells. If their clusters are consistent with those of all cells restricted to the samples, conclude that the clustering is stable. Divisive hierarchical spectral clustering is used. The mapping of the dendrogram to nested clusterings may be novel. Counts are transformed to points in Euclidean space. Positive affinities are defined for points that are k-nearest neighbors. The affinity equals the inverse of the distance between points. Ng, Jordan, and Weiss’ algorithm divides the points into two clusters. The normalized cut measures the clusters’ separation. Recursion generates a dendrogram. Set the length of the branch between a node and its daughters to the normalized cut. Nodes’ distances from the root define the mapping to nested clusterings. Analyze for all cells and multiple pairs of complementary samples. For a given number of clusters, compare each sample’s clustering and clusters with those of the full data set, providing stability measures. Results For three large data sets, this found clusterings compatible with published results, though with fewer clusters. Clusterings of two were judged to be stable. Conclusions It is feasible to identify stable clusterings of as many as 100,000 cells. Future research should explore using differential expression for validation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.