I have a small scRNA-Seq dataset (n = 357, inhibitory neurons). This set of cells is split almost evenly between two conditions (Case and Control). I would like to test for differential expression with MAST [1]. However, with such a small number of cells and high heterogeneity of different inhibitory neuron cell types, I am able to get only a single gene as differentially expressed (False Discovery Rate 0.05, Log Fold Change 0.15).

The best solution would be to enrich this subset of neurons in a new sample, however, are there other approaches that can be used to extract more information from this dataset? I'm not considering imputation at the moment since it seems to produce a large number of false positives [2].

  1. Finak, Greg, et al. "MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data." Genome biology 16.1 (2015): 1-13.
  2. Andrews, Tallulah S., and Martin Hemberg. "False signals induced by single-cell imputation." F1000Research 7 (2018).
  • $\begingroup$ Maybe your Case and Control cells are simply not that different transcriptionally? Have you examined their separation with t-SNE or similar? $\endgroup$ Jul 23 '20 at 19:28
  • $\begingroup$ @Chris_Rands good point, I think your suggestion explains the small log-Fold-Change. However, the Case is composed of cells with a specific heterozygous deletion, therefore I would expect to recover as differential at least the deleted genes. This is indeed what is validated in bulk RNASeq. Differently, in single-cell data, we noticed a drop in differential expression almost perfectly correlated with the decreasing number of cells. This is expected, but before running an additional sample with enriched cell types we wanted to look for existing techniques to recover part of the signal. $\endgroup$
    – gc5
    Jul 24 '20 at 0:57
  • $\begingroup$ @Chris_Rands to reply directly to your question, they are not separated on visualization projections such as UMAP or t-SNE. We don't expect, however, a very big separation since we expect the processes involved in the deletion to be more subtle than processes associated with cell-type identity. To conclude, when we have bigger clusters of cells, we are able to recover (positive control) the genes deleted in the Case. $\endgroup$
    – gc5
    Jul 24 '20 at 1:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.