Unexpected splitting of high-AF genomes into different secondary clusters in dRep

[Pranjal-Bioinfo]

Hi dRep team,

I’m encountering a possible issue with how dRep assigns secondary clusters even when pairwise ANI and alignment fraction (AF) values are high.

Here’s an example from my dataset:

SPAdes-CONCOCTRefined-1UOYPL_HR.28.fa,SPAdes-CONCOCTRefined-9TDUTH_S296.109.fa,0.964682,0.3767441860465116,13
SPAdes-CONCOCTRefined-9TDUTH_S296.109.fa,SPAdes-CONCOCTRefined-1UOYPL_HR.28.fa,0.969092,0.6694214876033058,13

Comparison Direction ANI (%) Coverage (%) Primary Cluster
1UOYPL_HR.28 → 9TDUTH_S296.109 96.47% 37.67% 13
9TDUTH_S296.109 → 1UOYPL_HR.28 96.91% 66.94% 13

These two genomes belong to the same primary cluster (13) and are connected with:

Despite this, dRep splits them into different secondary clusters (13_1 and 13_2),
Field 1UOYPL_HR.28 9TDUTH_S296.109
primary_cluster 13 13
secondary_cluster 13_1 13_2
centrality 0.05 0.05
ANI and Coverage pass (≥ 95%, ≥ 0.30) pass (≥ 95%, ≥ 0.30)

Is there a reason dRep's secondary clustering would behave this way even when the ANI/AF thresholds are satisfied?

Thanks for the great tool, and I’d really appreciate clarification!

[Pranjal-Bioinfo]

also i calculated manually for these two MAGs using fastANI

A → B: ANI=96.91%, AF=0.669
B → A: ANI=96.47%, AF=0.377

I used these parameters in drep command : -sa 0.95 -pa 0.9 -nc 0.30 -cm larger

[MrOlm]

Hi @Pranjal-Bioinfo - my guess is that this has to do with using "average" hierarchical clustering, which can make unexpected clusters sometimes. See here for details:

https://drep.readthedocs.io/en/latest/choosing_parameters.html#oddities-of-hierarchical-clustering

Best,
Matt

[Pranjal-Bioinfo]

Thanks @MrOlm it helped a lot.
But I have one more doubt, even if genome A and genome B has more than cutoff ANI, and nc is also more than cutoff from genome A to genome B but less when genome B is compared to genome A, in that case also drep puts them into different clusters. I am using --clusterAlg single.
What would you suggest ? I have metagenomes and I am generating SGBs.
Thanks a lot for help.