# 2. Ligand–receptor inference

**Language:** Python · **Input:** integrated `.h5ad` · **Output:** `{KO,WT}_lr_ready_custom.csv`

In this step, we will identify ligand–receptor interactions for the knockout (KO)
and wild-type (WT) samples separately using LIANA+'s CellPhoneDB implementation. First, we load the integrated data and the consensus resource before running the analysis.

:::{admonition} Add the Il1rn interaction manually
:class: important

The consensus resource does not contain the interaction **Il1rn → Il1r1_Il1rap**.
Because **Il1rn** is the antagonist ligand at the centre of this study, this interaction
must be added manually. Otherwise, the knockout effect cannot be scored.
:::

## Score interactions per condition

```python
import scanpy as sc, pandas as pd, liana as li
from liana.method import cellphonedb

adata = sc.read_h5ad(DATA / "bone_marrow_Il1rn_KOvsWT_harmony_integrated.h5ad")

CELLTYPE  = "cluster_names"
CONDITION = "Condition"

consensus_db = li.resource.select_resource("mouseconsensus")
custom_db = pd.concat(
    [consensus_db, pd.DataFrame([{"ligand": "Il1rn", "receptor": "Il1r1_Il1rap"}])],
    ignore_index=True,
)

adata.raw = adata
for cond in set(adata.obs[CONDITION]):
    lr = cellphonedb(
        adata[adata.obs[CONDITION] == cond],
        groupby=CELLTYPE,
        expr_prop=0.1,          # LR expression proportion threshold
        resource=custom_db,
        use_raw=False,
        inplace=False,
        verbose=True,
    )
    lr.to_csv(LR_OUT / f"{cond}_lr_liana_consensus.csv")
```

## Reshape for CrossTalkeR

CrossTalkeR expects a different column layout from the one produced by LIANA+.
The following code keeps only statistically significant interactions
(`cellphone_pvals <= 0.05`), renames the required columns, and writes one
CrossTalkeR-compatible file for each condition.

```python
import os

for f in os.listdir(LR_OUT):
    if not f.endswith("lr_liana_consensus.csv"):
        continue

    ev = pd.read_csv(LR_OUT / f)
    ev = ev.loc[:, ["ligand", "receptor_complex", "source", "target",
                    "lr_means", "cellphone_pvals"]]
    ev["type_gene_A"] = "Ligand"
    ev["type_gene_B"] = "Receptor"
    ev["gene_A"]  = ev["ligand"]
    ev["gene_B"]  = ev["receptor_complex"]
    ev["MeanLR"]  = ev["lr_means"]
    ev = ev.loc[ev["cellphone_pvals"] <= 0.05, :]

    cond = f[: f.find("_lr_")]
    ev.loc[:, ["source", "target", "type_gene_A", "type_gene_B",
               "gene_A", "gene_B", "MeanLR"]].to_csv(
        LR_OUT / f"{cond}_lr_ready.csv"
    )
```

The output files (`KO_lr_ready_custom.csv` and `WT_lr_ready_custom.csv`) are now
ready to be imported into CrossTalkeR.

In the next step, we perform TF activity predictions:
Next: [Regulon and TF activity](03_tf_activity.md)