# Workflow to integrate dataset from two different species

**URL:** <https://discourse.scverse.org/t/workflow-to-integrate-dataset-from-two-different-species/3432>\
**Category:** scvi-tools\
**Created:** [January 17, 2025, 12:56am UTC](https://discourse.scverse.org/t/workflow-to-integrate-dataset-from-two-different-species/3432 "2025-01-17T00:56:48Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Slack90](https://avatars.discourse-cdn.com/v4/letter/s/aeb1de/32.png) [@Slack90](https://discourse.scverse.org/u/Slack90)\
**Post date:** [January 17, 2025, 12:56am UTC](https://discourse.scverse.org/t/workflow-to-integrate-dataset-from-two-different-species/3432/1 "2025-01-17T00:56:48Z")

</div>

Hi,

I’d like to use scvi to integrate data from the same brain region in two different species. By using clustering and manual annotation I could identify that both datasets contain the same clusters (each cluster corresponding to a cell type). Now I would like to use scvi to integrate both dataset.  
So first I find the orthologue genes to both datasets, and then I subsample only the genes are orthologue between the two species. Then I do some filtering and normalization as follows:

```auto
    adata = filtered_anndata
    adata.var_names_make_unique()
    sc.pp.filter_cells(adata, min_genes=2) #get rid of cells with fewer than 200 genes
    sc.pp.filter_genes(adata, min_cells=1) #get rid of genes that are found in fewer than 3 cells
    adata.var['mt'] = adata.var_names.str.startswith('mt-') # annotate the group of mitochondrial genes as 'mt'
    sc.pp.calculate_qc_metrics(adata, qc_vars=['mt'], percent_top=None, log1p=False, inplace=True)
    upper_lim = np.quantile(adata.obs.n_genes_by_counts.values, .999)
    lower_lim = np.quantile(adata.obs.n_genes_by_counts.values, .001)
    adata = adata[(adata.obs.n_genes_by_counts < upper_lim) & (adata.obs.n_genes_by_counts > lower_lim)]
    adata = adata[adata.obs.pct_counts_mt < 20]
    sc.pp.normalize_total(adata, target_sum=1e4) #normalize every cell to 10,000 UMI
    sc.pp.log1p(adata) #change to log counts
    sc.pp.highly_variable_genes(adata, min_mean=0.0125, max_mean=3, min_disp=0.5) #these are default values
    adata.raw = adata #save raw data before processing values and further filtering
    adata = adata[:, adata.var.highly_variable] #filter highly variable
    sc.pp.regress_out(adata, ['total_counts', 'pct_counts_mt']) #Regress out effects of total counts per cell and the percentage of mitochondrial genes expressed
    sc.pp.scale(adata, max_value=10) #scale each gene to unit variance
    sc.tl.pca(adata, svd_solver='arpack')
    sc.pp.neighbors(adata, n_neighbors=10, n_pcs=30)
    sc.tl.leiden(adata, resolution = 0.25)
    sc.tl.umap(adata)

```

Then I concatenate the two datasets and prep for scvi:

```auto
concatenated_anndata = concatenated_anndata

concatenated_anndata.obs.groupby('dataset').count()
# sc.pp.filter_genes(concatenated_anndata, min_counts=3)
concatenated_anndata.layers["counts"] = concatenated_anndata.X.copy() # preserve counts
sc.pp.normalize_total(concatenated_anndata, target_sum=1e4)
sc.pp.log1p(concatenated_anndata)
concatenated_anndata.raw = concatenated_anndata # freeze the state in `.raw`

scvi.model.SCVI.setup_anndata(
    concatenated_anndata,
    layer="counts",
    categorical_covariate_keys=["dataset"])

model = scvi.model.SCVI(concatenated_anndata)

```

However the results I get from the integration are pretty terrible:

 ![image](https://canada1.discourse-cdn.com/flex035/uploads/forum11/original/2X/a/a4007c10dfe777341ff6f880dcaad95457c9e444.png)

I thought one possible reason might be that the highly variable genes selected (using the parameters I have above) lead to a very low number of common genes between the two dataset (from 10.000 of orthologues, down to ~300). So I tried to skip altogether the filtering of highly variable genes, but results look still pretty bad. So I am wondering if there is something wrong with the order of the steps I am following for this pipeline (whether I should do the selection of highly variable genes before finding the orthologues?) or the pipeline in general. Does anyone have any suggestions?

---

<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:** [January 21, 2025, 11:18am UTC](https://discourse.scverse.org/t/workflow-to-integrate-dataset-from-two-different-species/3432/2 "2025-01-21T11:18:40Z")

</div>

Hi,  
I dont see the full pipeline of your analysis here,  
From what I see you should do the preprocessing above without scaling the data **after** datasets were concatenated with the same genes

See our documentation for similar tutorials