What is the best way to address the question of doublets and multiplets in a single cell RNA seq data set?

Question

I have attached a histogram plot of the number of genes per cell in a single cell RNA seq data set of lung endothelial cells. I do not find a bimodal or multimodal distribution of the number of genes per cell within any particular condition. Doublets/multiplets should have an aberrantly high number of gene expression and if many, can lead to a bi or multimodal histogram plot of the number of genes detected per cell. Can I conclude that I have an insignificant presence of doublets or multiplets in my data set from this plot and hence continue with downstream analyses?

I am reluctant to use automatic doublet-finding algorithms because they have a high possibility to remove cells in a transitional differentiating phase. I also think that the default threshold of 2500 genes per cell given by Seurat might sometimes lead to the loss of good cells depending on the cell type.

Apart from trajectory analyses, are there any other analyses on single cell RNA seq data that are sensitve to doublets/multiplets?

Thanks in advance for your help.

Answer 1

Indeed there's no satisfactory doublet/multiplet identification methods as of 2020.

The most reliable way to remove doublets is still manual removal of doublet subclusters guided by marker genes.

Not only trajectory analysis can be impacted by doublets, but clustering and embedding can also be affected for obvious reasons.