I have tried MaAsLin to find association of microbial genus with meta-characters. I got a lot of significantly associated genus, including some seems to be noises. May i know is it meaningful if i filter the results based on the value of coefficients? Like if i set a threshold for the absolute value of coefficients and only consider genus with coefficient larger than this threshold. If so, do you have any recommendations about the threshold? By the way, i have also set the threshold for q-values for filtering.
Thanks in advance.
Hi @Qiuyu-dot - filtering or subsetting differential analysis results based on a combination of effect size estimates and qvalues is standard practice, but I am not sure if there is a recommended threshold since it can vary from study to study.
Thanks for the reply. I am not sure whether I understand MaAsLin correctly. Will the abundance influence the effect size? If the relative abundance of one species is low, will there also be higher chance that its effect size could also be small (if there is significant effect)? I am worrying that since I have already set the minimum abundance for features before running MaAsLin, will it be too strict to set threshold for effect size again?
Hi @Qiuyu-dot - great question. The independent pre-filtering you are referring to is done to ensure that some low-variable features (which have a low chance of being statistically significant) are filtered out before modeling with the hope that they do not have sufficient variance that can be explained by the model. In the latter case, you have already restricted yourself to the ‘set of most variable features’ and you would like to further rank them by their effect size estimates. Since the purpose of pre-filtering is orthogonal to the purpose of post-filtering, I would not be concerned by the impact of one on another.
When you talked about effect size, did you mean the absolute value of coefficient?
Yeah, I mean the absolute value of coefficients. Is there any recommendation about the threshold of this value?