Interpreting ANCOM and ANCOM-BC results

Hi @jwdebelius

I am opening this conversation again to ask you some questions.
I've done the Ancom Volcano Plot on phyla level and this is my result
l2-ancom-treatment.qzv (471.8 KB)

I also tried at the other levels, but no significant results appeared.
How do you interpret this result?

Is there a way to show this result with another graph? I saw that this is not the common way to represent this result in papers.

Thanks again for your help :grinning:

Hi @Linda_Abenaim, I'd recommend using ANCOM-BC rather than ANCOM for your differential abundance analysis, and then visualizing the results with da-barplot, which should be a lot more straight-forward to interpret than the volcano plot.


Hi @Linda_Abenaim,

To add on to @gregcaporaso's excellent advice, I think its worth considering a couple possibilties.

  1. There aren't differences at phylum level. This wouldn't suprise me. Members of hte same phyla can have very little in common and still be grouped together. For example, humans, sharks, and lancet worms are all part of the same phylum (phylum chordata) but beyond having spinal cords and bilateral symmetry, there isn't much we share from an ecological perspective! Pat Schloss did some interesting work a while ago which suggested family was a sweet spot for colorectal cancer research. I dont know enough about insect guts to know if this will hold true or if you'll need to go deeper, but family tends to be the limit of my taxonomic levels.

  2. Statistical power due to dimensionality. If you're seeing a difference in beta diveristy but not detecting differentailly abundant taxa, it may simply be an issue of statistical power. ANCOM is conservative and often has issues with small sample sizes. This can be managed by filtering your data. The good news is that ANCOM-BC which Greg recommended does osme filtering for you, so you should have fewer concerns about low abundance/low prevelance features decreasing your power.

  3. Low statistical power due to sample size. Sometimes, a sample size that is big enough for a difference in beta diveristy isn't big enough for a difference in differential abundance, or you're underpowered to detect certain samples. I dont know about your system or your expected effect size, so I'm not sure where you fall. In my system of choice, I worry at fewer than 50 samples, but I also deal with free living organisms who have a tendency to do what they want. :person_cartwheeling:

  4. There is no signal. It's possible there just isn't a difference between your groups, or there may not be a difference you can detect between your groups. This might be indicated by a lack of signal in beta diversity, especially when you test with something like adonis. Unfortunately, there's a publication bias toward positive results, but it doesnt mean your negative results dont add information.



Thank you so much for your answer, @jwdebelius!
I will also try the suggestion of @gregcaporaso.
I obtained significant results during L2 and L4.
I'm not so good at interpreting the Ancom Volcano plot, but from these two, I can understand that:
-in L2: actinobacteriota is abundant in treated and bacteriodata in control

-in L4: These are my results, so Corynobacteriales is abundant in the treated diet compared to the control diet.

Did I well understand?
l2-ancom-treatment.qzv (471.8 KB)
l4-ancom-treatment.qzv (473.1 KB)

However, I think that I have to restart my analysis; I think to denoise to much my sequence because I have very low features.
I'm repeating my analysis with denoising (F trunc len 245, R trunc len 245) instead of (F trunc len 177, R trunc len 192).
I had a very strange result in my old feature table and stats:
stats-dada2.qzv (1.2 MB)
table-dada2.qzv (449.6 KB)

Now my results are these:
stats-dada2-new.qzv (1.2 MB)
table-dada2-new.qzv (554.1 KB)

what do you think?

my amplicon region was V3-V4

thanks again

Hi @Linda_Abenaim,

Im going to again recommend that you use ANCOM-BC because I think it will be easier to interpret. About all I'm getting out of your current results is that you have a low complexity community and that's making your W values hard to intrepet. I think what both your volcano plot and results here are saying is that you dont have a statitistically significant difference in your results.

In my last post, I talked about some reasons you might get that result. Do any of those seem like they apply to your work?



Thank you @jwdebelius!
I will try ANCOM BC!

Iā€™m repeating my analysis from dada2, because I noted that I was mistake to filter and I filtered to much and at the end I obtained a feature table exported with very few feature.

So I think that this could be a problem also about this ANCOM VOLCANO plot results.

I will try again my analysis with right trim and turn of dada2 and then I will elaborate again my taxonomy analysis! I will let you know :muscle:t2:

1 Like

Hi @jwdebelius!

I tried again my anaysis from dada2 and now my results on stats dada2 are better.
I tried ANCOM-BC but I don't understad very well this visualization.

these are my command, i don't know if i forgot something important in these command:

qiime composition ancombc
--i-table table.qza
--m-metadata-file sample-metadata.txt
--p-formula treatment
--o-differentials dataloaf.qza

qiime composition da-barplot
--i-data taxonomysilva/dataloaf.qza
--o-visualization taxonomysilva/dataloaf-barplot.qzv

these are my sample-metadata:
sample-metadata-grouped.txt (790 Bytes)

I don't understand if treattmentt--> the second t is for t= treatment in my sample-metadata.txt.
How i can obtain results for treated (t) and control (c)?
Can I see the differences of abbundance from the two treatment in this visualization?
dataloaf-barplot.qzv (322.8 KB)

I replicate also Ancom volcano plot.
These are my result:
l5-ancom-treatment.qzv (485.5 KB)
l6-ancom-treatment.qzv (487.3 KB)

Could you give me your opinion?

Thanks for your help! This is my first time and i have a lot of doubt on this taxonomic analysis.

Hi everyone! grinning:

Could someone help me to interpret my ancom-bc and dabarplot results?

here my sample-metadata:
sample-metadata-grouped.txt (790 Bytes)
I have two diets (control and treated) and I would like to see the difference in the abundance of bacterial community between these two diets.

I've done ancom-bc at 5 level:
ancombc-l5.qzv (319.1 KB)
and then dabarplot:
dabarplot-l5.qzv (319.4 KB)

If I well understood in dabarplot I can find the blue bars that are referred to treated diet and organge ones to control diet. I understood that dabarplot represents the prevalence and abundance of bacterial communities in these two diet. So for example, "enterococcaceae" are present in the treated diet (blue one) in greater abundance than the control (where I cannot find this bacterial family).
Did I interpret these results well?
What about the significance?

Thank you so much

Hi @Linda_Abenaim,

I split this off and moved it to a new topic because it was getting away from hte other discussion.
I also combined this post with your other post. Please don't double post the same content; the mod team is all volunteers and double posting wastes time.

That said, let's talk about results

ANCOM-BC is essentially running a linear regression under the hood. I'd recommend reading about regression with categorical variables if you're unfamiliar with the technique; its a good foundation for a lot fo the analysis here. In a categoricla regression, we have to set some group as our reference. In most models, its usually the group that comes first. So, here, "C" is the reference group, and you're comparing "T".

In this case, these bars describe the relationship between the treated group and the control group. Blue means higher in the treated group, orange is lower in the treated group compared to the control. (Positive and negative changes based on your intercept). I think you always have to talk about things in relative terms and be clear aout what youre reference group is.

The features shown in the waterfall plot are the ones that are statistically significant, but youd have to read the documentation or poke around the forum for more details about what the specific parameters are to identify them.



Thank you so much @jwdebelius!
Iā€™m so sorry for the other discussion!

So to summarise, the families that are less abundant in the treated diet are more abundant in the control diet? Do you think that can I show only this in a thesis for example, or also for control?

Do you think that i have to put * where there are significativity?

Hi @Linda_Abenaim
Yes, Everything in ANCOM-BC is a comparison. So a Family is enriched in the control diet compared to the treated diet.

On the QIIME 2 forum, its important that your work is your own. I personally this this question is bordering asking our moderators to make your conclusions about you work for you. Our user support category is to ask questions about understanding methods, but our moderators can't provide support on what should be included in your thesis.

Generally this is a common practice, but overall its a personal decision for how you want to display your conclusions of this analysis.


Thank you so much for your answer @cherman2!

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.