# Maaslin Analysis with Non-clinical data and Small samples

**URL:** <https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537>\
**Category:** MaAsLin\
**Created:** [October 1, 2025, 2:47am UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537 "2025-10-01T02:47:07Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![ADAM\_BIN\_AZMIHAN](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/adam_bin_azmihan/32/3433_2.png) [@ADAM\_BIN\_AZMIHAN](https://forum.biobakery.org/u/ADAM_BIN_AZMIHAN)\
**Post date:** [October 1, 2025, 2:47am UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/1 "2025-10-01T02:47:07Z")

</div>

Hello, i am a postgraduate studying rats, Currently, I’m working on microbiome for my final project. I would like to ask advice on how to best set up MaAsLin for my analysis.

This is my experimental design:  
Animals obtained from LAFAM are divided into 4 groups n=7/groups. Each groups has different treatments saline, D-galactose, Soymilk and postbiotics. So right now, i am analysing for the effects of the treatment to the gut microbiome. So basically i want to see the change in microbial feature corresponding to the treatment.

i have done the analysis but the adjusted p value is not significant. so i am wondering if my experimental design is fit to be analyse using Maaslin, because as far as i understand the min n = should be 10 times the predictor.

---

<div class="post-metadata">

**Author:** ![nearinj](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/nearinj/32/2135_2.png) [@nearinj](https://forum.biobakery.org/u/nearinj)\
**Post date:** [October 1, 2025, 4:54pm UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/2 "2025-10-01T16:54:58Z")

</div>

Hi @ADAM_BIN_AZMIHAN,

Your study does look like it has smaller sample sizes. I would take a look at your model and see if every covariate you included in the model is something you want to correct for in the output p-values.

Depending on your interest you may be able to just correct the p-values from treatment alone (without the other covariates) which should increase the power of your study. Again this is only recommended if you truly are only interested in the treatment outcome.

Cheers,  
Jacob

---

<div class="post-metadata">

**Author:** ![ADAM\_BIN\_AZMIHAN](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/adam_bin_azmihan/32/3433_2.png) [@ADAM\_BIN\_AZMIHAN](https://forum.biobakery.org/u/ADAM_BIN_AZMIHAN)\
**Post date:** [October 2, 2025, 1:01am UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/3 "2025-10-02T01:01:56Z")

</div>

Hi Jacob,

Thanks a lot for your helpful suggestion! That makes sense — I’ll re-examine the covariates in my model and consider whether I should restrict the correction to treatment alone, since that’s the main effect I’m interested in.

I also wanted to ask: in the MaAsLin 3 paper they mention a kind of _post-hoc filtering_, where small but statistically significant prevalence effects (with coefficients \<1) were considered likely compositional artifacts, and filtering those out before recalculating FDR helped restore high precision. Do you think applying a similar approach would be valid in a small-sample setting? And if so, should I re-run the FDR correction only after filtering for stronger coefficients, or apply filtering after significance is already determined?

Thanks again for the guidance

---

<div class="post-metadata">

**Author:** ![nearinj](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/nearinj/32/2135_2.png) [@nearinj](https://forum.biobakery.org/u/nearinj)\
**Post date:** [October 6, 2025, 5:11pm UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/4 "2025-10-06T17:11:34Z")

</div>

@ADAM_BIN_AZMIHAN

In the manuscript we filtered those features after applying FDR not before and so I’m not sure it will help in your situation. This is because you don’t know the effect size you are looking for a-prior and so an effect size less then 1 may actually be something interesting in your dataset (and not just a compositional effect).

Cheers,  
Jacob

---

<div class="post-metadata">

**Author:** ![ADAM\_BIN\_AZMIHAN](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/adam_bin_azmihan/32/3433_2.png) [@ADAM\_BIN\_AZMIHAN](https://forum.biobakery.org/u/ADAM_BIN_AZMIHAN)\
**Post date:** [October 7, 2025, 1:38am UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/5 "2025-10-07T01:38:40Z")

</div>

Got it, thanks Jacob! I’ll focus on the effect sizes and shifts then — looks like my data might be a bit underpowered. Really appreciate your help!

Adam

---

<div class="post-metadata">

**Author:** ![Han\_Ming\_Gan](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/han_ming_gan/32/3350_2.png) [@Han\_Ming\_Gan](https://forum.biobakery.org/u/Han_Ming_Gan)\
**Post date:** [October 7, 2025, 11:10am UTC](https://forum.biobakery.org/t/maaslin-analysis-with-non-clinical-data-and-small-samples/8537/6 "2025-10-07T11:10:05Z")

</div>

Hi Adam,

I assume you’re in Malaysia . Try give a buzz to patriot biotech - they maybe can be of help.