I plan on collecting some drop-seq data from brain tissue (from mice and humans). Using only the 75-85% of genes that have 1:1 orthology between mice and humans, I'd like to cluster the cells by transcriptomic similarity and get a dendrogram to use for cell type classification. There are tons of tools for processing and clustering single cell RNA-seq data (https://www.scrna-tools.org/tools), but I get the impression that most of them were not designed with Drop-seq in mind. Anyone have any insight into the pros and cons of the best tools for Drop-seq analysis?
-
$\begingroup$ What are the particularities in Drop-seq that would require a different tool/method to analyze it? (It is the first time I heard about drop-seq and I am not totally sure I understood the difference between drop-seq and sc-seq) $\endgroup$– llrsOct 16, 2017 at 14:33
-
$\begingroup$ Compared with other scRNA-seq techniques (like Smart-seq), Drop-seq sacrifices read counts for cell counts. That's the big difference. It's also very cost-effective and has been increasing in popularity since it came out in 2015. cell.com/molecular-cell/abstract/S1097-2765(17)30049-7 $\endgroup$– Eric A BrennerOct 16, 2017 at 14:54
4 Answers
This article uses the freely available R package dropbead for filtering and then Seurat to perform a principal component analysis that groups together affine transcriptomes. It could be what you are looking for.
-
$\begingroup$ I've seen that paper and read the abstract, but I didn't know about Dropbead. Thank for the info! $\endgroup$ Oct 17, 2017 at 15:00
-
$\begingroup$ Note that there isn't any license for the package. It might not be safe to use it. (Although I am sure that if you contact the authors they will grant you permission and correct this :) $\endgroup$– llrsOct 17, 2017 at 15:24
-
$\begingroup$ The DESCRIPTION file states GNU GPL3, but yes it would be good to have a LICENSE file $\endgroup$– PeterJun 23, 2018 at 13:26
SC3 is part of the Bioconductor library for R. It is based on PCA and spectral dimensionality reductions and utilizes k-means. In the introductory paper the authors perform analysis on a Drop-seq dataset of more than 44,000 cells. SC3 outputs several useful plots, including a consensus matrix, expression panel, and a marker genes heatmap. All of these utilize the hierarchical clustering.
I can't speak from personal experience but the package seems very user friendly. Being a Bioconductor package means it has lots of support and excellent documentation. I am getting ready to use it myself for a small Drop-seq dataset of olfactory epithelial cells.
SC3: consensus clustering of single-cell RNA-seq data : Nature Methods : Nature Research. Available at: https://www.nature.com/nmeth/journal/v14/n5/full/nmeth.4236.html. (Accessed: 31st October 2017)
You can try a hierarchical clustering tool HGC (hierarchical graph clustering).
You may get cell population at different hierarchy by choosing multiple cutting heights. See the paper for more details.
The current version of Seurat and Scanpy can both handle Drop-seq data. If you want to include more low-reads cells you can loosen the filtering thresholds in either of these packages. These two packages really can handle both low-read-high-cell data and high-read-low-cell data.
This ENCODE paper used Seurat to process both C1 Fluidigm data (>4000 genes per cell) and 10X data (>1000 genes per cell only).
This human limb paper used Seurat to process Drop-seq data, although they didn't annotate the cell types well. I used Seurat to reprocess their data and it works much better than what the paper presents.
I guess the key is parameter tuning.