Does an OTU interfere in another?

[Renan Soares]

Hi! I have some doubts about LefSe analysis. Could somebody please help me?

 

I have made some tests with the template table available on the website. I tested removing the subclasses and I noticed that if I remove some OTUs, the results of the others change.

 

For example, if use the whole table with no subclasses, the results are the following:

 

Bacteria,Bacteroidetes	5,220474			-
Bacteria,Firmicutes	5,356907			-
Bacteria,Actinobacteria	5,52609	non_mucosal	5,470808	0,009023
Bacteria,Proteobacteria	4,730767			-
Bacteria,Firmicutes,Clostridia	4,614166			-
Bacteria	5,688297

 

So, in this case, the only discriminative OTU is Actinobacteria.

 

If I remove the OTU “Bacteria”, results are the following:

Firmicutes	5,649788			-
Clostridia	4,8932			-
Bacteroidetes	5,495121			-
Proteobacteria	5,023754			-
Actinobacteria	5,817617	non_mucosal	5,762311	0,009023

 

And if keep only “Actinobacteria”  and “Bacterioidetes”, results are the following:

 

Bacteroidetes	5,827436	mucosal	5,759768	0,016294
Actinobacteria	5,973105	non_mucosal	5,759768	0,016294

 

So, in this case, “Bacterioidetes” is now a discriminant OTU, what it was not in the first analysis.

 

I’d like to understand the following:

1 – Why one OTU interfere in another? In my first understanding, the software tests each OTU individually.

2 – If I use “no subclasses”, Wilcoxon test is performed? Or just Kruskal-Wallis?

3 – What do the results on the second column in the results table mean? I understand that the third is LDA Score and the fourth the p value.

4 – If have many OTUs, but would like to test just a few is it possible or correct to do the test if one OUT interfere in the others?

5 – If I have more than 2 classes, a discriminat OTU would indicate an OTU that is discriminant comparing it with “both other 2 classes” or with “at least 1” class?

Thank you so much

 

Renan

[Renan Soares]

Hi! Could someone please help me with these questions?

[Nicola Segata]

Hi Renan,

 with the default settings, LEfSe uses relative abundance. So it normalize the sum of OTUs to a fixed value. This means that if you remove an OTU which was not equally abundant in all samples, then the results can change after the normalization.

thanks

Nicola

[Renan Soares]

Hi, Nicola!

 

Thank you so much for your clear and fast answer. It was so clarifying and helped me to understand my results!

 

About question 2, for what I’ve read, Wilcoxon test is not performed when no subclasses are present, right?

 

The second column on the results output is the highest average of the class. Changing the OTUs this value changes a little sometimes, maybe because of the fact you mentioned in your answer, right?.

  

Regards,

 

Renan

[Nicola Segata]

Hi Renan,

  the Wilcoxon test is performed also when no subclasses are present, but at least for the case of two classes this is equivalent to the first Kruskal-Wallis test.

Yes, I think that the highest average in the classes are also affected by the renormalization after OTU removal.

cheers

Nicola

[jfg]

Dear Nicola, 

   Can I suggest this post (or one similar, e.g. a quick FAQ) is stickied to the top of the forum, as the question "does one OTU interfere in another" is one that crops up again and again?

sincerely,

jfg