Quantitative assessment of single-cell RNA-seq clustering with CONCORDEX
01:49 Share

Link to bioRxiv paper: http://biorxiv.org/cgi/content/short/2023.06.28.546949v1?rss=1

Authors: Jackson, K. C., Booeshaghi, A. S., Galvez-Merchan, A., Moses, L., Chari, T., Pachter, L.

Abstract: Many single cell RNA-sequencing (scRNA-seq) data analysis workflows rely on methods that embed and visualize the properties of a k-nearest neighbor (kNN) graph in two-dimensions. These visualizations are typically combined with categorical labels assigned to individual data points and can support a range of analysis tasks, despite the fact that these embeddings are known to distort the local and global properties of the graph. Rather than relying on a two-dimensional visualization, we introduce a method for quantitatively assessing the concordance between a set of labels and the k-nearest neighbor graph. Our method, called CONCORDEX, computes for each node the fraction of neighbors with the same or different label, and compares the result to the mean obtained with random labeling of the graph. CONCORDEX can be used for any categorical label and can be interpreted via an intuitive heatmap visualization. We demonstrate its utility for assessment of clustering results and "integration". Since CONCORDEX can be used to directly visualize properties of a kNN graph, we also use CONCORDEX to evaluate how well two-dimensional embeddings capture the local and global structure of the underlying graph. We have made CONCORDEX available as a Python-based command line tool (https://github.com/pachterlab/concordex) and as a software package in Bioconductor (https://bioconductor.org/packages/concordexR).

Copy rights belong to original authors. Visit the link for more info

Podcast created by Paper Player, LLC