ABSTRACT
The clustering of single cell RNA-sequencing (scRNA-seq) data enables the classification of cell types, and the development of integration algorithms has enabled the tracing of altered-from-baseline transcriptomes to their respective cell type origins. Here, we developed an algorithm that removes sources of noise from scRNA-seq datasets and improves the accuracy of their integration. We found that this improvement of integration is further increased by an algorithm we developed for auto-determination of optimal weight-assignment to anchors. As a proof-of-concept, we show that our algorithm determined the type-origin of the 17% of injured retinal ganglion cells (RGCs) that an existing integration algorithm could not identify. As we found that the majority of the originally unassigned cells belonged to only several out of 45 RGC types, a consequent change in the proportions of the surviving types resulted in a moderately different ranking of resiliency to injury. This updated standard for RGC type resilience ranking improved the understanding of RGC type biology and the ranking-dependent prediction of genes for therapeutic neuroprotective targeting. Additional bioinformatic analyses revealed single-gene cluster-markers for RGC types and contributed new insights into the global characteristics of RGC types and how axonal injury affects them, showing how the extent of variance between transcriptomes of RGC types increases differentially during maturation and after injury. We also show, for the first time, that specific global properties of the transcriptome can predict the resilience to injury of certain cell types. Furthermore, we developed a website for cluster-by-cluster comparison of gene expression between uninjured and injured RGC types. The publicly available R library package, CellTools, for denoising and improving the integration of scRNA-seq datasets, as well as for associating cell types with the global properties of the transcriptome, will assist scRNA-seq studies of developing, injured, or diseased cell types across biological fields.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
New data was added to increase impact.
https://health.uconn.edu/neuroregeneration-lab/subtypes-gene-browser