Alignment threshold and secondary cluster assignments - not sure it's working properly?

[mcmahon-uw]

Hi Matt!

I'm using dRep v3.4.5.

I've been using the compare command to look at secondary genomes clusters that all belong to the same primary cluster. I noticed that some are being assigned to their own secondary cluster even though their ANI is above my threshold. However, their alignment FRACTION with many other genomes is only 0.12-0.4 because they are SAGs and quite incomplete. I thought this was happening because the minimum coverage threshold default was set to be very low, but the 0.1 default should still group those genomes based more on their ANI. I tried rerunning with the coverage threshold explicitly set to 0.1, but got the same results.

Here is my code:
dRep compare -p 24 -g ~/Mendota_genomes/1_dRep/starting_genomes/*fna -sa 0.96 --cov_thresh 0.1 dRep

Here is a snippet of my output:

26Sep2015rr0052-bin.134.fna,SAG_2739367632.fna,0.978604,0.1875,1 27Jul2012rr0045-bin.200.fna,SAG_2739367632.fna,0.969617,0.25,1 MAGv2_3300020483-bin.4.fna,SAG_2739367632.fna,0.977835,0.19377162629757785,

It seems like 26Sep2015rr0052-bin.134.fna,SAG_2739367632, and MAGv2_3300020483-bin.4 should all be in the same cluster. Is it splitting them because of the way cluster-wide ANI is calculated (i.e. average?). Or some other reason?

Love all your tools and they are so accessible. Thank you for your efforts!

[MrOlm]

Hi Trina,

Thanks for the kind words! A couple of things that could be going on:

1. Is it possible that the genomes are being split into different primary clusters (based on Mash)? That can happen with incomplete genomes, and a potential workaround would just be to run with --SkipMash (assuming you don't have so many genomes that skipping primary will take forever)

2. It could very well be due to the average ANI clustering, especially if some genomes are going below that 0.1 threshold (which can mess with things). A way to get around this would be to add --clusterAlg single, which ensures that any genomes above that -sa threshold are in the same final cluster (though it can lead to larger than normal clusters).

3. A way that I sometimes find helpful to diagnose issues like this is by looking at the Secondary_clustering_dendrograms figure, which explicitly shows how this final clustering is done

Happy to follow up if this doesn't answer your question!

Best,
Matt

[mcmahon-uw]

Sweet! Thanks for the speedy response.
I actually grabbed these out of ~ 17k genome collection, based only on the MASH clustering. I got 30 genomes from that cluster, which I'm then putting into the secondary clustering.

I played around with the algorithm setting. "Single" actually only gave me one big secondary cluster again. But "median" gave me two, and that matches better what I was expecting (from working with these genomes in other ways). I might keep playing and see what I get with the other ones, just out of curiosity.

Thank you!!