Here we present VISION, a module that can sit downstream of other common analyses such as clustering, dimensionality reduction, and trajectory inference. Specifically, VISION aids in the interpretation of scRNA-seq data, with or without predetermined labels, or stratifications of the data (e.g. clusterings) using the notion of cell-cell similarity maps (as interpreted from some latent space) and biological signatures (functional sets of genes that can be obtained online from, for example, MSigDB). Finally, VISION can evaluate the effect of cell- level meta data, such as library quality, batch, clinical information, or additional experimental readouts (e.g., protein levels from a Cite-Seq experiment). Importantly, the use of VISION can greatly facilitate collaborative projects, as it offers a low- latency interactive report for the end- user, which can be hosted online and viewed on a web browser without the need for installing developer-grade software.

For more information, refer to our preprint on biorxiv: Functional Interpretation of Single-Cell Similarity Maps

We recommend installing VISION via github using devtools:

require(devtools)
install_github("YosefLab/VISION", build_vignettes=T)

We recommend using the build_vignettes argument to view vignettes from R (use the vignette() function).

VISION generally follows the same pipeline from iteration to iteration, where minor differences can be specified via the various parameters in a VISION object. On a typical VISION run:

• For large datasets, or if the user so chooses, micropools are computed - grouping similar cells together to reduce the complexity of the analysis.
• If a latent space is not specified, PCA is performed and the top 30 components are retained.
• A KNN graph is constructed from the latent space, named the cell-cell similarity map
• Signature scores are computed using the expression matrix
• Signature local “consistencies” on the cell-cell similarity map are computed using the Geary-C statistic, an auto-correlation statistic.
• An interactive web-based report is generated that can be used to explore and interpret the dataset.

You can refer to the vignettes to run VISION. To note, there is an extra vignette detailing how to properly interface with Dynverse for incorporating VISION into your trajectory inference pipeline.

Click here for an example output report of ~9,000 CBMC's sequenced with the CITE-seq protocol