Starlng: STability Analysis of coReguLated "Nests" of Genes

Single-cell assays (both single and multi-omics) represent the state-of-the-art for detailed characterisation of cell phenotypes, a task frequently initialised with the identification of groups of genes linked to well-defined, stable subpopulations of cells. Gene expression covariation patterns are simplified precursors of Gene Regulatory Networks (GRNs), and play a crucial role in inferring causality and dynamics within GRNs. However, current gene clustering methods struggle to capture spatial organisation and relationship dynamics along developmental trajectories, while remaining vulnerable to noise and algorithmic stochasticity.

Expanding on the ClustAssess framework, which shifted the cluster-centric assessment of populations of cells to an element-centric perspective, we propose to assess the covariation induced by gene graph clustering. The Starlng (abbrev) framework provides an optimised, interactive environment for the identification of the filtered modules. The pipeline identifies the stable configuration of clustering parameters and robust gene modules (nests of genes), followed by the removal of the outlier clusters based on the pseudotime variance. We illustrate Starlng on the identification of reliable, reproducible communities of genes, as well as reveal unbiased hierarchical relationships between higher-level cell clusters and more granular subgroups of genes.

Starlng is implemented in R and freely available at https://github.com/Core-Bioinformatics/Starlng .

Supplementary data are available at Bioinformatics online.