interpretation of log-fold change in ANCOM-BC

Hi, I got a more basic question about ANCOM-BC. For the results I got, I would like to know the amplitude of log-fold change, i.e. to know the base of the log indicated in this method. After some search, I found ANCOM-BC probably applies centered log-ratio transformation which is probably exponential based. So, I will interpret a log-fold change of 1.1 as being more abundant at a ratio of e^1.1 than control. Is it correct??

Thanks for the reply

1 Like

Hi @b87401116,

Great question! ANCOM-BC utilizes the log() method in R, which by default is the natural log (i.e. log base e). If you're interested, you can see where in the original R method takes the log here (which we've wrapped as a method in q2-composition). Hope this helps! Cheers :lizard:

3 Likes

Hi @lizgehret,
Thanks a lot! I would like to confirm that 'centered' log transformation is applied in this method also. After centered log transformation, values greater than 1 may not be greater than zero after transformation, because a certain shift of the center value is applied. Am I right?

Hi @b87401116,

A CLR transform is not used in ANCOM-BC - you are most likely thinking of ANCOM, which does use a CLR transformation.

ANCOM-BC accounts for sampling fraction by introducing a sample-specific offset term in a linear regression framework that is estimated from the observed data. The offset term serves as the bias correction, and the linear regression framework in log scale is analogous to log-ratio transformation to deal with the compositionality of microbiome data.

I would recommend checking out the ANCOM-BC paper, it goes into great detail on the methods used - and is helpful in understanding how this method differs from ANCOM.

Hope this helps! Cheers :lizard:

2 Likes

Hi @lizgehret
Thank you for the clarification. I think I did mess them up.

Best

1 Like