# Differential Expression using Scanpy

**URL:** <https://discourse.scverse.org/t/differential-expression-using-scanpy/1058>\
**Category:** scanpy\
**Created:** [January 12, 2023, 1:56pm UTC](https://discourse.scverse.org/t/differential-expression-using-scanpy/1058 "2023-01-12T13:56:49Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![PaulJonasJost](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/pauljonasjost/32/499_2.png) [@PaulJonasJost](https://discourse.scverse.org/u/PaulJonasJost)\
**Post date:** [January 12, 2023, 1:56pm UTC](https://discourse.scverse.org/t/differential-expression-using-scanpy/1058/1 "2023-01-12T13:56:49Z")

</div>

Hello!

I am trying to do a differential expression analysis on three different clusters using `tl.rank_genes_groups`. I do have more than three clusters but only want to compare cluster 1 (in the following named C1) with Cluster 2 ( C2) and Cluster 3 (C3) respectively.

Now I have two questions regarding this:

1. What is the correct code?  
Looking at the API, I thought of 2 ways, the first would be to call `tl.rank_genes_groups` twice using `groups` to filter for `C1, C2` and `C1, C3`, i.e.

```auto
sc.tl.rank_genes_groups(
    adata = adata,
    groupby = "clusters",
    groups=["C1", "C2"],
    method="wilcoxon",
    corr_method="benjamini-hochberg"
)

```

The problem I have here is that I do not know how exactly `reference` is working together with `groups`. The default there is “rest”, but does that refer to “the rest of all other groups in clusters” or “the rest of all other groups which are in the groups argument”. The first code would be for the latter, while for the former I would probably have to set the reference, i.e.

```auto
sc.tl.rank_genes_groups(
    adata = adata,
    groupby = "clusters",
    groups=["C2", "C3"],
    reference = "C1"
    method="wilcoxon",
    corr_method="benjamini-hochberg"
)

```

Which one would be the correct one here?

1. Visualization

When I visualise the result using `pl.rank_genes_groups_dotplot`, I get all clusters on my y-axis, but I would only want C1, C2, C3. How can I do that? The `groups` argument seems to only influence the genes shown.

Thank you very much for this great tool and in advance already for the help 🙂

---

<div class="post-metadata">

**Author:** ![mxposed](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/mxposed/32/27_2.png) [@mxposed](https://discourse.scverse.org/u/mxposed)\
**Post date:** [January 18, 2023, 12:37am UTC](https://discourse.scverse.org/t/differential-expression-using-scanpy/1058/2 "2023-01-18T00:37:21Z")

</div>

Hi

1. According to the [code](https://github.com/scverse/scanpy/blob/d26be443373549f26226de367f0213f153556915/scanpy/tools/_rank_genes_groups.py#L158) I think `reference="rest"` will only take the groups you provided. So the best way to find marker genes for C1, C2, C3 between themselves is to just pass `groups=["C1", "C2", "C3"]`. This will compare C1 with C2+C3, C2 with C1+C3 etc.
2. Looking at the code again, unfortunately I don’t see an option for this subsetted plotting. A clumsy workaround would be to compute gene expression and size yourself and pass it to `pl.rank_genes_groups_dotplot` as `dot_color_df` and `dot_size_df` arguments. See [here](https://github.com/scverse/scanpy/blob/d26be443373549f26226de367f0213f153556915/scanpy/plotting/_dotplot.py#L172)