# Variability in melonnpan result

**URL:** <https://forum.biobakery.org/t/variability-in-melonnpan-result/4829>\
**Category:** MelonnPan\
**Created:** [February 21, 2023, 8:06am UTC](https://forum.biobakery.org/t/variability-in-melonnpan-result/4829 "2023-02-21T08:06:34Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![paramartha\_banerjee](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/paramartha_banerjee/32/1858_2.png) [@paramartha\_banerjee](https://forum.biobakery.org/u/paramartha_banerjee)\
**Post date:** [February 21, 2023, 8:06am UTC](https://forum.biobakery.org/t/variability-in-melonnpan-result/4829/1 "2023-02-21T08:06:34Z")

</div>

Hello!! I have trained the melonnpan model using the input metabolomic and gene family files provided in the tutorial. However now when I am attempting to test the melonnpan predict workflow using the same gene family files used for training I am getting different relative abundances of predicted metabolites (it is supposed to predict similar(if not the same ) relative abundance values of metabolites that we used for training purposes)  
what might be the reason for such variability in results?  
I have attached the input metabolomic file and also the output obtained after melonnpan predict  
[metabouputpredict.csv](https://forum.biobakery.org/uploads/short-url/IqXyGlbcH0V3zmogOmP7yjtgKf.csv) (142.3 KB)  
[metabinputtraining.csv](https://forum.biobakery.org/uploads/short-url/aZ69peTmvy4ObLZmqfBa3DHEdwY.csv) (148.1 KB)

---

<div class="post-metadata">

**Author:** ![himel.mallick](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/himel.mallick/32/1757_2.png) [@himel.mallick](https://forum.biobakery.org/u/himel.mallick)\
**Post date:** [February 24, 2023, 2:05am UTC](https://forum.biobakery.org/t/variability-in-melonnpan-result/4829/2 "2023-02-24T02:05:54Z")

</div>

Hi @paramartha_banerjee - the predicted values are supposed to be correlated with the measured values assuming that the corresponding metabolite is well-predicted but they are not supposed to be identical (unless the correlation between the measured and predicted value is 1 which is rarely the case). Likewise, this variation is expected as the prediction accuracy is not always perfect. Hope this makes sense.

---

<div class="post-metadata">

**Author:** ![paramartha\_banerjee](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/paramartha_banerjee/32/1858_2.png) [@paramartha\_banerjee](https://forum.biobakery.org/u/paramartha_banerjee)\
**Post date:** [February 24, 2023, 11:49am UTC](https://forum.biobakery.org/t/variability-in-melonnpan-result/4829/3 "2023-02-24T11:49:38Z")

</div>

Thanks for clarifying!! I have one more question  
Is predicting metabolites using HUMAnN3 generated gene family data with the default trained model possible?

---

<div class="post-metadata">

**Author:** ![himel.mallick](https://yyz1.discourse-cdn.com/flex027/user_avatar/forum.biobakery.org/himel.mallick/32/1757_2.png) [@himel.mallick](https://forum.biobakery.org/u/himel.mallick)\
**Post date:** [February 28, 2023, 2:27am UTC](https://forum.biobakery.org/t/variability-in-melonnpan-result/4829/4 "2023-02-28T02:27:17Z")

</div>

It might be possible with the caveat that the default model was trained with version 2 but in general, as long as you have sufficient matching UniRef90s across versions, you can use the default model.

Thanks a bunch,  
Himel