# Unstable result

**URL:** <https://discourse.scverse.org/t/unstable-result/1090>\
**Category:** scvi-tools\
**Tags:** scvi\
**Created:** [February 1, 2023, 4:23pm UTC](https://discourse.scverse.org/t/unstable-result/1090 "2023-02-01T16:23:40Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![GwangWooKim](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/gwangwookim/32/512_2.png) [@GwangWooKim](https://discourse.scverse.org/u/GwangWooKim)\
**Post date:** [February 1, 2023, 4:23pm UTC](https://discourse.scverse.org/t/unstable-result/1090/1 "2023-02-01T16:23:40Z")

</div>

Hello, I’m going to analyze the single-cell RNA sequencing data using the scVI framework.  
After training the scVI model, I found that the result of model.differential\_expression is unstable (specifically, it varies from seed to seed).  
I’ve been trying to solve this problem, but I still haven’t.  
Please tell me the expected cause.  
Thank you in advance.

---

<div class="post-metadata">

**Author:** ![maheshworpaudel5001](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/maheshworpaudel5001/32/1601_2.png) [@maheshworpaudel5001](https://discourse.scverse.org/u/maheshworpaudel5001)\
**Post date:** [March 27, 2026, 3:07pm UTC](https://discourse.scverse.org/t/unstable-result/1090/2 "2026-03-27T15:07:48Z")

</div>

I am using scvi version 1.4.1 and I am facing the same problem. Any help would be really appreciated.

---

<div class="post-metadata">

**Author:** ![ori-kron-wis](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/ori-kron-wis/32/1099_2.png) [@ori-kron-wis](https://discourse.scverse.org/u/ori-kron-wis)\
**Post date:** [March 27, 2026, 3:54pm UTC](https://discourse.scverse.org/t/unstable-result/1090/3 "2026-03-27T15:54:08Z")

</div>

> [@GwangWooKim](#):
>
> Hello, I’m going to analyze the single-cell RNA sequencing data using the scVI framework.  
> After training the scVI model, I found that the result of model.differential\_expression is unstable (specifically, it varies from seed to seed).  
> I’ve been trying to solve this problem, but I still haven’t.  
> Please tell me the expected cause.  
> Thank you in advance.

The DE analysis is based on sampling, so there will be variability between runs.

But given your model converged (check it) and that there is a real biological signal in the data (not just noise, do HVG) - you can run more MC samples to expect more stable results between seeds. Is it version-dependent?