Is it worth to explore doing single cell analysis outside of Seurat, or is the Seurat package pretty comprehensive for basic analyses? Is it worth it to look into other approaches outside of Seurat for quality control, dimensionality reduction, visualization... and what would the advantages be? Genuinely intrigued by this!
Absolutely! Seurat is valuable for many types of analysis but I personally found it to be quite a blackbox because either the vignette code works or not, and there is little documentation available on what the functions are actually doing internally. I also found the general accessability of the Seurat object cumbersome and unintuitive, therefore I went all-in onto Bioconductor for single-cell analysis, see for an excellent read: http://bioconductor.org/books/release/OSCA/
OSCA has excellent documentation, lots of example code and background information plus developers and maintainers are outstandingly responsive on both support.bioconductor.org and Slack. The Seurat developers are active at Github as well but many issues are either unanswered or answered vaguely or not to satisfaction (at least for me). It is one single group maintaining the whole package so I do not blame them.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Indeed, there is also scanpy, which is quite well-regarded and well-supported.
The Bioconductor ecosystem is nice because there are both very robust, stable analysis packages and more cutting-edge tools as well. Generally, these are interoperable given the common data structures used by Bioconductor and tend to be updated quickly due to the more modular nature of the single cell analysis ecosystem. It's not a single group writing everything from start to finish (though there are a few extremely prolific actors, e.g. Aaron Lun), so support responsibilities are generally more well-distributed and it's generally easier to find answers as to both why and how various tasks are performed. OSCA is definitely a worthwhile read even if you choose to stick with Seurat - it provides clear explanations of various processing steps and why they're necessary (or not in some cases).
I think the biggest upside of Bioconductor, beyond its large user base and active developers, is the interoperability. The SingleCellExperiment format (which is the workhorse storage format or "data type" for single-cell analysis) can be accepted by basically any Bioc tool that is remotely related to single-cell analysis. There are converters of course to switch between Seurat and SCE plus between any format and Anndata format that scanpy is using, so one could use tools from multiple universes without much trouble.
Agree fully. I initially went all in on Seurat as well when creating my web app (docs.dseqr.com) but switched completely to Bioconductor after discovering the OSCA handbook. I have since integrated some functionality from Seurat such as reference based mapping with the excellent Azimuth references (https://azimuth.hubmapconsortium.org/).
Thank you! I will look into OSCA.