# 3. Regulon and TF activity

**Language:** Python · **Input:** integrated `.h5ad` ·
**Output:** `mouse_dorothea_AB.csv`, `decoupler_dorotheaAB_results_ulm.csv`

In this step, we will estimate transcription factor (TF) activity for every
cell using the DoRothEA regulon and decoupleR's univariate linear model (ULM).
The resulting TF activity scores are used for differential analysis with IntraTalker.

## Load the DoRothEA regulon

IntraTalkerpy includes the complete mouse DoRothEA regulon. We recommend using
only the highest-confidence interactions (confidence levels **A** and **B**),
so filter the regulon before estimating TF activity.

:::{admonition} Use only confidence levels A and B
:class: important

The bundled regulon contains all confidence levels (A–D). Throughout this
tutorial, only confidence levels **A** and **B** are used because they provide
the most reliable transcription factor–target interactions. Using the complete
regulon will produce different TF activity estimates and downstream results.

You can also use custom regulon databases instead of the DoRothEA regulons provided with the package.
:::

```python
from importlib.resources import files
import pandas as pd
import intratalkerpy

reg = pd.read_csv(
    files("intratalkerpy.tf.data").joinpath("mouse_dorothea_reg.csv"))

reg = pd.read_csv(regulon_path)
reg = reg.loc[reg["confidence"].isin(["A", "B"]), ["source", "target", "weight"]]

reg.to_csv(TF_OUT / "mouse_dorothea_AB.csv", index=False)

```

The filtered regulon is saved for reuse in the downstream R workflow.

## Estimate transcription factor activity

Next, we estimate transcription factor activity for every cell using decoupleR's
univariate linear model (ULM). This produces one activity score for every
transcription factor in every cell.

```python
import decoupler as dc

dc.mt.ulm(
    data=adata,
    net=reg,
    verbose=True,
    raw=False,
)

estimates = adata.obsm["score_ulm"]
estimates.to_csv(TF_OUT / "decoupler_dorotheaAB_results_ulm.csv")
```

The output is a cell-by-transcription factor activity matrix. Differential
testing of these activity scores (KO versus WT within each cell type) is
performed in the next step using R.

:::{admonition} Handoff: Python → R
:class: note

The IntraTalker and CrossTalkeR analysis of the workflow is implemented in R. Before continuing, make sure
you have the following files:

- `bone_marrow_Il1rn_KOvsWT_harmony_integrated.rds`
- `mouse_dorothea_AB.csv`
- `decoupler_dorotheaAB_results_ulm.csv`
- `KO_lr_ready.csv`
- `WT_lr_ready.csv`

Once these files are available, you can start the R workflow.
:::

**Next:** [IntraTalker + CrossTalkeR2](04_intratalker_crosstalker.md)