02-integration.qmd

---
title: "Organoid integration"
---

# Introduction

In this document we will perform integration of the two organoid samples.

```{python}
import lamindb as ln
import lnschema_bionty as lb
import scanpy as sc
import anndata
import scvi

lb.settings.species = "human"

transform = ln.Transform(name="Organoid_integration")
ln.track(transform)
```

# Load Data

## Organoid 123

```{python}
org123_file = ln.File.filter(key="Organoid123_qc.h5ad").one()
org123 = org123_file.load()
org123
```

## Organoid 4

```{python}
org4_file = ln.File.filter(key="Organoid4_qc.h5ad").one()
org4 = org4_file.load()
org4
```

# Unintegrated

## Concatenate

```{python}
orgs = anndata.concat([org123, org4], merge="same")
orgs.var["gene_symbols"] = org123.var.loc[orgs.var_names.to_list()]["gene_symbols"]
```

## UMAP

```{python}
sc.pp.highly_variable_genes(
    orgs,
    flavor="cell_ranger",
    n_top_genes=2000,
    batch_key="Dataset"
)
sc.pp.pca(orgs)
sc.pp.neighbors(orgs)
sc.tl.umap(orgs)

sc.pl.umap(orgs, color="Dataset")
```

# Prepare

```{python}
scvi_adata = orgs[:, orgs.var.highly_variable].copy()

scvi.model.SCVI.setup_anndata(
    scvi_adata,
    layer="counts",
    batch_key="Dataset",
)

scvi_adata
```

# Train

```{python}
model = scvi.model.SCVI(scvi_adata)
print(model)
model.train(max_epochs=100)
```

# Integrated

```{python}
orgs.obsm["X_scVI"] = model.get_latent_representation()
orgs.obsm["X_umap_unintegrated"] = orgs.obsm["X_umap"].copy()

sc.pp.neighbors(orgs, use_rep="X_scVI")
sc.tl.umap(orgs)

sc.pl.umap(orgs, color="Dataset")
```

# Save

```{python}
orgs_file = ln.File.from_anndata(
    orgs,
    field=lb.Gene.ensembl_gene_id,
    key="Organoids_integrated.h5ad"
)
orgs_file.save()
```