Fig. 1: Schematic of the mechanisms behind q-diffusion.
From: q-Diffusion leverages the full dimensionality of gene coexpression in single-cell transcriptomics
When comparing two cells, the kernel fundamentally values expression differences that occur in many genes concurrently. It can enter and augment several common analyses: a q-Diffusion facilitates nuanced phenotype resolution via community detection, as with the second case study in this paper. b q-Diffusion can regularize gene expression program (GEP) estimators like nonnegative matrix factorization (NMF), to promote statistical enrichment of gene ontologies (first and third case studies). c Recent spatial scRNAseq modalities present a new opportunity for macro-segmentation based on cellular transcriptomics, like in the brain (third case study). We present a local distributional segmentation (LDS) algorithm that relies on q-diffusion applied to maximum mean discrepancy (MMD), an established kernel-based statistic.
