erinyoung
commented
5 months ago Roary is no longer maintained. (https://github.com/sanger-pathogens/Roary). I ran into many issues with Roary that several other people had also run into, such as having samples randomly being dropped from the analysis.
When it came time to update Grandeur, I knew I needed to replace Roary with something currently maintained. The options I chose to experiment with were:
Admittedly, I didn't have a lot of time for this comparison. So in addition to my sample set (a cluster of seven Serratia samples), I relied on the literature. The two most influential papers were file:///Users/eriny/Downloads/s13059-020-02090-4.pdf and https://www.microbiologyresearch.org/content/journal/mgen/10.1099/mgen.0.000690.
All of these tools tout themselves as being a roary replacement, but they are all slightly different from roary and from each other.
Here's a figure that may or may not be helpful: https://link.springer.com/article/10.1186/s13059-020-02090-4/figures/2
In summary, for core genome analysis, panaroo does identify more pairwise SNPs that Roary.
The default arguments for panaroo in Grandeur are --clean-mode strict --remove-invalid-genes, which is supposedly the best way to remove possible contamination. As observed in the file that you sent to me, the total number of genes accepted for each isolate in the analysis is smaller (i.e. 2017-30-24 goes from 3,775 genes identified by roary to 3,702 genes identified by panaroo). Panaroo is more selective as to what is actually a gene. Genes are then matched to other isolates through other methods. Roary used blast, and panaroo uses cd-hit. It looks like panaroo is using genes in the core genome that roary does not match together. Your Version v3.6.20231219 (roary) has a core genome of 3,187 genes, while version 4+ has 3,255 genes. I imagine these genes hold most of your additional SNPs.
This unfortunately, is going to be a similar story with both ppanggolin and pirate as well.
Panaroo was used for Grandeur due to its results being basically the same in my sample set and ease of use. I had issues running pirate with singularity (perl issues ftw!), and ppanggolin is much more complicated to use.
I think there is a way to adjust panaroo to run more like roary (https://gthlab.au/panaroo/#/gettingstarted/params), but the sample set that I was using doesn't look like a good fit for this. Are the samples in your comparison on the SRA? It looks like it would be perfect for fine tuning some arguments.
whottel
Hello,
I went back through an old cluster analysis to see what differences in interpretation would potentially arise in using with an older grandeur version that uses roary versus newer versions that are now using panaroo. Please find a summary excel file with this comparison: Grandeur_Version_Matrix_Comparisons_Issue.xlsx
This comparison includes sequences that were part of a 2017 P. mirabilis outbreak. Grandeur with roary identified 0-3 SNPs among these outbreak sequences. With other versions of grandeur that use panaroo, two sequences seem to form their own subcluster and differ by >200 SNPs from the other members of the outbreak cluster. Surprisingly, 2017-A and 2017-D are from the same sample but again with panaroo differ by hundreds of SNPs. Also, included is a third-party refMLST analysis, which generally agrees with the roary SNP matrix as far as interpretation.