We used AutoGeneS29 (v.1.0, https://github.com/theislab/AutoGeneS) to deconvolve the tomo-seq data. As input, we used the single-cell data constrained to the 5,000 most variable genes using scanpy75 (pp.highly_variable_genes). AutoGeneS was then used to select a total of 400 informative genes from among the highly variable ones that differentiated the cell types. The highly variable genes were selected based on normalized dispersion. The proportions were inferred using nonnegative least squares76 (scipy.optimize.nnls) based on the cellular mean expression of the informative genes. For each bulk sample, negative proportions were set to zero and the rest were normalized to sum to one.
Do you have any questions about this protocol?
Post your question to gather feedback from the community. We will also invite the authors of this article to respond.