Ancom shows very low number of taxa

Hello

I have more than 6000 taxa in the taxonomy.qzv, but most of them present in only one or two sample, i have 22 samples. when i ran ancom on taxonomic level7, i found only 110 taxa that are differentially abundant.
Notice:i filtered the frequency table to contain only feature that greater than 50frequency and present in 4 samples and more( --p-min-features 50 ,–p- min-samples 4)
Is it normal to obtain only 100 taxa in ancom?

Also, another strange result, i get only one feature when i run ancom without the taxa collapse command, why?


Thanks for the great team

Hi @arar,

The number of taxa you get in ANCOM depends a lot on your experimental parameters. However, given that you have 22 samples, it’s actually quite high to me! If I were you, I probably would filter my data more to remove features that are only present in a few samples (although your sample size is quite small to begin with.) The things you’re seeing are likely related to rare (low prevelance) organisms. So, I’d remove those and try again. (I might set my threshhold at 2-4 if I were you.) This relates to your second question.

This may have to do with opposite behavior of close related organisms in the same genus, or, again, the sparsity of your data in your very small sample size. I trust this result much more than I trust your per-feature results.

Best,
Justine

2 Likes

Good

I already filtered sequences and table and get 720 taxa for 720 feature, only 110 of them appear to be differentialy abundant with ancom.
Should i rerun alpha,beta diversity and beta group significant commands again on the filtered sequences and table or i can depend on the result of them before filtration?

1 Like

Hi @arar,

I tend to check that my filtering hasn’t majorly altered my community by calculating beta diversity on my filtered data and then using a mantel test. I look for a mantel R ≥ 0.95 as a measure of similarity, given that Ive often had to drop my rarefaction depth and that can lead to differences.
But, beyond that, I dont typically recalculate.

Best,
Justine

2 Likes

Ok i will try mantel test as a measurement of similarity
Thank you very much

1 Like