### Install Cell2net from PyPI

Source: https://github.com/pinellolab/cell2net/blob/main/docs/install.md

Use this command to install the latest stable version of cell2net from the Python Package Index. This is the recommended method for most users.

```bash
pip install cell2net
```

--------------------------------

### Install Cell2net from GitHub

Source: https://github.com/pinellolab/cell2net/blob/main/docs/install.md

Install the latest development version of cell2net directly from its GitHub repository. This is useful for accessing the newest features or bug fixes before they are released on PyPI.

```bash
pip install git+https://github.com/pinellolab/cell2net.git
```

--------------------------------

### Start Training for ZHX3

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the ZHX3 gene.

```text
Output:
Training model for gene: ZHX3

```

--------------------------------

### Setup Directories and Paths

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Defines the input path for multiome data and the output directory for training artifacts. Ensures the output directory exists.

```python
input_data = "./02_prepare_data/mdata.h5mu"
out_dir = './03_train'

os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Start Training for AL035458.2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the AL035458.2 gene.

```text
Output:
Training model for gene: AL035458.2

```

--------------------------------

### Setup File Paths and Output Directory

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/05_tf_to_gene.ipynb

Configure input and output directories for TF attribution analysis. Ensure the output directory exists.

```python
data_dir = "./02_prepare_data/mdata.h5mu"
in_dir = "./03_train_cell2net"
out_dir = "./05_to_gene"

os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Start Training for GPCPD1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the GPCPD1 gene.

```text
Output:
Training model for gene: GPCPD1

```

--------------------------------

### Setup Output Directories

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Creates necessary directories for organizing training outputs, including model checkpoints, plots, and predictions. The `exist_ok=True` argument prevents errors if directories already exist.

```python
os.makedirs(f"{out_dir}/model", exist_ok=True)
os.makedirs(f"{out_dir}/plot", exist_ok=True)
os.makedirs(f"{out_dir}/prediction", exist_ok=True)
```

--------------------------------

### Start Training for SULF2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the SULF2 gene.

```text
Output:
Training model for gene: SULF2

```

--------------------------------

### Setup Output Directory

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Creates a directory to store processed and standardized results for evaluation. The `exist_ok=True` argument prevents an error if the directory already exists.

```python
out_dir = "./05_process_p2g"

os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Setup Output Directory

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Creates the necessary output directories for storing trained models, plots, and predictions. The `exist_ok=True` argument prevents errors if directories already exist.

```python
out_dir = './03_train_cell2net'

os.makedirs(out_dir, exist_ok=True)
os.makedirs(f"{out_dir}/model", exist_ok=True)
os.makedirs(f"{out_dir}/plot", exist_ok=True)
os.makedirs(f"{out_dir}/prediction", exist_ok=True)
```

--------------------------------

### Start Training for AL110114.1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Initiates the training process for a specific gene. This output indicates the start of model training.

```text
Output:
Training model for gene: AL110114.1

```

--------------------------------

### Import Libraries and Set Options

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Imports necessary libraries for data processing and sets mudata options. Ensure these libraries are installed before running.

```python
import os
import pandas as pd
import mudata as md
import numpy as np

md.set_options(pull_on_update=False)
```

--------------------------------

### Start Training for TMX4

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the TMX4 gene.

```text
Output:
Training model for gene: TMX4

```

--------------------------------

### Start Training for MAFB

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the MAFB gene.

```text
Output:
Training model for gene: MAFB

```

--------------------------------

### Setup File Paths for Correlation Analysis

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/04_get_correlation.ipynb

Defines the input paths for the multiome dataset and trained Cell2Net models, and specifies the output directory for correlation analysis results. Creates the output directory if it does not exist.

```python
input_data = "./02_prepare_data/mdata.h5mu"
in_dir = './03_train_cell2net'
out_dir = './04_get_correlation'

os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Import Libraries and Set Options

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Imports necessary libraries for data manipulation, modeling, and visualization, and sets global options for the mudata library. Ensure these libraries are installed before running.

```python
import warnings
warnings.filterwarnings("ignore")
import os
import mudata as md
import cell2net as cn
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
md.set_options(pull_on_update=False)
import torch
```

--------------------------------

### Cell2Net Training Start for BACH1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, BACH1.

```text
Output:
Training model for gene: BACH1

```

--------------------------------

### Start Training for B4GALT5

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the B4GALT5 gene. This output signifies the beginning of model training.

```text
Output:
Training model for gene: B4GALT5

```

--------------------------------

### Cell2Net Training Start for EVA1C

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, EVA1C.

```text
Output:
Training model for gene: EVA1C

```

--------------------------------

### Start Training for CST3

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the CST3 gene, including initial data split information.

```text
Output:
Training model for gene: CST3
2025-11-21 02:43:01 INFO     Training size is provided: 0.8
2025-11-21 02:43:01 INFO     Split the data for training and validation

```

--------------------------------

### Cell2Net Training Start for AP000317.1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, AP000317.1.

```text
Output:
Training model for gene: AP000317.1

```

--------------------------------

### Cell2Net Training Start for CTSZ

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, CTSZ.

```text
Output:
Training model for gene: CTSZ

```

--------------------------------

### Start Training for RNF24

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the RNF24 gene.

```text
Output:
Training model for gene: RNF24

```

--------------------------------

### Import Libraries and Set Options

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/misc/01_create_mudata.ipynb

Imports necessary libraries and sets Muon options for efficient data handling. This is a common setup for MuData workflows.

```python
import warnings
warnings.filterwarnings("ignore")

import mudata as md
import muon as mu
import scanpy as sc

md.set_options(pull_on_update=False)
```

--------------------------------

### Start Training for ZNF341-AS1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the ZNF341-AS1 gene.

```text
Output:
Training model for gene: ZNF341-AS1

```

--------------------------------

### Install Cell2net in Development Mode

Source: https://github.com/pinellolab/cell2net/blob/main/docs/install.md

Install cell2net in editable mode for development. This method is suitable for contributors or users who plan to modify the source code, as changes will be reflected immediately without reinstallation.

```bash
git clone https://github.com/pinellolab/cell2net.git
cd cell2net
pip install -e .
```

--------------------------------

### Cell2Net Training Start for CHODL

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, CHODL.

```text
Output:
Training model for gene: CHODL

```

--------------------------------

### Import Libraries and Warnings

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Imports necessary Python libraries and configures warning filters for the project. This setup is standard for data analysis and machine learning tasks.

```python
import warnings
warnings.filterwarnings("ignore")

import numpy as np
from scipy import stats
import torch
import cell2net as cn
import mudata as md
import scanpy as sc
import matplotlib.pyplot as plt
import seaborn as sns

md.set_options(pull_on_update=False)
```

--------------------------------

### Start Training for SIGLEC1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of model training for the SIGLEC1 gene.

```text
Output:
Training model for gene: SIGLEC1

```

--------------------------------

### Cell2Net Training Start for PMEPA1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, PMEPA1.

```text
Output:
Training model for gene: PMEPA1

```

--------------------------------

### Start Training for SIRPB2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Initiates the training process for the SIRPB2 gene. This output signifies the beginning of model training.

```text
Output:
Training model for gene: SIRPB2

```

--------------------------------

### Start Training for C20orf194

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the initiation of the training process for the C20orf194 gene.

```text
Output:
Training model for gene: C20orf194

```

--------------------------------

### Cell2Net Training Start for ITSN1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, ITSN1.

```text
Output:
Training model for gene: ITSN1

```

--------------------------------

### Check Cell2Net Version

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Retrieves and displays the installed version of the cell2net library.

```python
cn.__version__
```

--------------------------------

### Cell2Net Training Start for ZBTB46

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, ZBTB46.

```text
Output:
Training model for gene: ZBTB46

```

--------------------------------

### Cell2Net Training Start for AP001347.1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, AP001347.1.

```text
Output:
Training model for gene: AP001347.1

```

--------------------------------

### Cell2Net Training Start for BCAS1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, BCAS1.

```text
Output:
Training model for gene: BCAS1

```

--------------------------------

### Load Required R Libraries

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/06_evaluate_p2g.ipynb

Loads necessary libraries for data manipulation, visualization, and analysis in R. Ensure these packages are installed before running.

```r
suppressMessages(library(cicero))
suppressMessages(library(ggplot2))
suppressMessages(library(cowplot))
suppressMessages(library(gridExtra))
suppressMessages(library(dplyr))
```

--------------------------------

### Cell2Net Training Start for AL109930.1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, AL109930.1.

```text
Output:
Training model for gene: AL109930.1

```

--------------------------------

### Cell2Net Training Output

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

This output indicates the start of training for a specific gene. It is part of the iterative training process for each gene in the dataset.

```text
Output:
Training model for gene: KCNJ15

```

```text
Output:
Training model for gene: BACE2

```

```text
Output:
Training model for gene: MX2

```

```text
Output:
Training model for gene: MX1

```

```text
Output:
Training model for gene: TRPM2

```

```text
Output:
Training model for gene: PCBP3

```

```text
Output:
Training model for gene: COL6A2

```

```text
Output:
Training model for gene: SAMSN1

```

```text
Output:
Training model for gene: NRIP1

```

```text
Output:
Training model for gene: AF130417.1

```

```text
Output:
Training model for gene: APP

```

```text
Output:
Training model for gene: ADAMTS5

```

```text
Output:
Training model for gene: AF165147.1

```

```text
Output:
Training model for gene: TIAM1

```

```text
Output:
Training model for gene: LINC00159

```

--------------------------------

### Start Training for SIRPD

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the commencement of model training for the SIRPD gene.

```text
Output:
Training model for gene: SIRPD

```

--------------------------------

### Cell2Net Training Start for MIR155HG

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, MIR155HG.

```text
Output:
Training model for gene: MIR155HG

```

--------------------------------

### Cell2Net Training Start for LINC01684

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, LINC01684.

```text
Output:
Training model for gene: LINC01684

```

--------------------------------

### Create Output Directory

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/02_prepare_data.ipynb

Create a directory to store the processed dataset and other output files. The 'exist_ok=True' argument prevents an error if the directory already exists.

```python
out_dir = "./02_prepare_data"
os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Cell2Net Training Start for MIR99AHG

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Indicates the start of the training process for a specific gene, MIR99AHG.

```text
Output:
Training model for gene: MIR99AHG

```

--------------------------------

### Load Pre-trained Sequence Encoder

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Initializes a pre-trained sequence encoder from a specified file path. Loads the state dictionary, mapping it to the CPU.

```python
pretrained_model_path = "./pretrained_seq2acc.pth"
pretrained_state_dict = torch.load(pretrained_model_path, map_location="cpu")

```

--------------------------------

### Define Data Directories

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/04_peak_to_gene.ipynb

Configures input and output directory paths for data preprocessing, model training, and peak-to-gene analysis. Ensures the output directory exists.

```python
data_dir = "./02_prepare_data/mdata.h5mu"
in_dir = "./03_train"
out_dir = "./04_peak_to_gene"

os.makedirs(out_dir, exist_ok=True)
```

--------------------------------

### Display Prepared Training Data Head

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Displays the first few rows of the prepared training data DataFrame, showing sequence and accessibility information.

```python
df_seq.head()
```

--------------------------------

### Select Essential Columns for Training

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Prepares the training and validation dataframes by selecting only the 'sequence' and 'acc' columns and resetting the index.

```python
df_train = df_train[['sequence', 'acc']].reset_index(drop=True)
df_valid = df_valid[['sequence', 'acc']].reset_index(drop=True)
```

--------------------------------

### Import Libraries and Set Options

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Imports necessary libraries for data manipulation, modeling, and visualization. Sets Mudata options to disable pull on update for efficiency.

```python
import warnings
warnings.filterwarnings("ignore")

import os
import mudata as md
import cell2net as cn
import torch
import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
md.set_options(pull_on_update=False)
```

--------------------------------

### Import Libraries and Set Options

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/04_peak_to_gene.ipynb

Imports necessary Python libraries for data manipulation, modeling, and Cell2Net. Sets global options for mudata and filters warnings.

```python
import os
import warnings

warnings.filterwarnings("ignore")

import numpy as np
import mudata as md
import pandas as pd
from tqdm import tqdm
import cell2net as cn
md.set_options(pull_on_update=False)
```

--------------------------------

### Read MuData Object

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/misc/01_create_mudata.ipynb

Demonstrates the basic command to read a MuData object. Ensure the MuData object is correctly formatted before reading.

```python
mu.read
```

--------------------------------

### Training Completion for AL035458.2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Reports the successful training of the AL035458.2 model, including the best epoch, validation correlation, and confirmation of the training data split.

```text
Output:
2025-11-21 02:45:22 INFO     Training finished
2025-11-21 02:45:22 INFO     Find best model at epoch 22 with valid correation  0.452
2025-11-21 02:45:23 INFO     Training size is provided: 0.8
2025-11-21 02:45:23 INFO     Split the data for training and validation

```

--------------------------------

### Display SCENT DataFrame Head

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Shows the first two rows of the SCENT peak-to-gene prediction DataFrame to inspect its structure and content.

```python
df_scent.head(n=2)
```

--------------------------------

### Prepare Training Data for Sequence Encoder

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Adds peak regions with specified window size and extracts DNA sequences from a reference genome. Requires 'cell2net' imported as 'cn' and 'numpy' as 'np'. Ensure the reference FASTA file path is correct.

```python
cn.pp.add_peaks(mdata, mod_name='atac', peak_len=256)
cn.pp.add_dna_sequence(mdata, ref_fasta='../../../../data/refdata-gex-GRCh38-2020-A/fasta/genome.fa')
```

--------------------------------

### Training Completion for SIRPB2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Logs the completion of training for SIRPB2, detailing the best epoch and validation correlation, and confirming the data split for training and validation.

```text
Output:
2025-11-21 02:37:53 INFO     Training finished
2025-11-21 02:37:53 INFO     Find best model at epoch 0 with valid correation  0.843
2025-11-21 02:37:55 INFO     Training size is provided: 0.8
2025-11-21 02:37:55 INFO     Split the data for training and validation

```

--------------------------------

### Load and Preprocess Multiome Data

Source: https://github.com/pinellolab/cell2net/blob/main/README.md

Load multiome data using MuData and perform basic preprocessing steps including adding peak information and DNA sequences. Ensure you have the 'genome.fa' file available.

```python
import cell2net as cn
import mudata as md

# Load example data
mdata = md.read_h5mu("path/to/multiome_data.h5mu")

# Basic preprocessing
cn.pp.add_peaks(mdata, mod_name='atac')
cn.pp.add_dna_sequence(mdata, ref_fasta='genome.fa')

# Create and train a model
model = cn.tl.Cell2Net(mdata, gene='GENE_OF_INTEREST')
model.train()
```

--------------------------------

### Import Libraries for Cell2Net Analysis

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/05_tf_to_gene.ipynb

Imports necessary Python libraries for data manipulation, Cell2Net analysis, and visualization. Sets options for mudata. Warnings are filtered for cleaner output.

```python
import warnings
warnings.filterwarnings("ignore")

import os
import numpy as np
import mudata as md
import cell2net as cn
from tqdm import tqdm
import pandas as pd
md.set_options(pull_on_update=False)
```

--------------------------------

### Import Libraries and Filter Warnings

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/06_tf_activity.ipynb

Imports necessary libraries for data manipulation, analysis, and Cell2Net. Filters out warnings to keep the output clean.

```python
import warnings

warnings.filterwarnings("ignore")

import pandas as pd
import scanpy as sc
import mudata as md
import pandas as pd
import cell2net as cn
md.set_options(pull_on_update=False)
```

--------------------------------

### Define Input Directory

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/06_evaluate_p2g.ipynb

Set the path to the directory containing processed peak-to-gene predictions from a previous step.

```r
in_dir <- "./05_process_p2g"
```

--------------------------------

### Load Pretrained Sequence Encoder Weights

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Loads pretrained weights for the sequence encoder from a specified file. This is useful for transfer learning to improve model convergence and performance.

```python
pretrained_model_path = "./pretrained_seq2acc.pth"
pretrained_state_dict = torch.load(pretrained_model_path, map_location="cpu")
```

--------------------------------

### Training Completion for SULF2

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Confirms the completion of training for SULF2, providing the best epoch, validation correlation, and details on the training data split.

```text
Output:
2025-11-21 02:47:38 INFO     Training finished
2025-11-21 02:47:38 INFO     Find best model at epoch 1 with valid correation  0.872
2025-11-21 02:47:39 INFO     Training size is provided: 0.8
2025-11-21 02:47:39 INFO     Split the data for training and validation

```

--------------------------------

### Training Completion for ZNF341-AS1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Confirms the completion of training for ZNF341-AS1, providing the best epoch, validation correlation, and details on the training data split.

```text
Output:
2025-11-21 02:44:33 INFO     Training finished
2025-11-21 02:44:33 INFO     Find best model at epoch 30 with valid correation  0.392
2025-11-21 02:44:34 INFO     Training size is provided: 0.8
2025-11-21 02:44:34 INFO     Split the data for training and validation

```

--------------------------------

### Training Completion for ZHX3

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Reports the successful training of the ZHX3 model, including the best epoch, validation correlation, and confirmation of the training data split.

```text
Output:
2025-11-21 02:46:53 INFO     Training finished
2025-11-21 02:46:53 INFO     Find best model at epoch 7 with valid correation  0.569
2025-11-21 02:46:54 INFO     Training size is provided: 0.8
2025-11-21 02:46:54 INFO     Split the data for training and validation

```

--------------------------------

### Initialize Seq2Acc Model

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Initializes a Seq2Acc model with a specified input sequence length (peak_len) and dropout rate for regularization. This model uses CNNs to learn sequence patterns for predicting chromatin accessibility.

```python
model = cn.pd.model.Seq2Acc(peak_len=256, dropout_rate=0.25)
```

--------------------------------

### Display Signac DataFrame Head

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Shows the first two rows of the Signac peak-to-gene prediction DataFrame to inspect its structure and content.

```python
df_signac.head(n=2)
```

--------------------------------

### Visualize RNA UMAP

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Generates a UMAP visualization for RNA data. Ensure 'scanpy' is imported as 'sc'.

```python
sc.pl.umap(mdata["rna"])
```

--------------------------------

### Training Completion for TMX4

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Confirms the completion of training for TMX4, providing the best epoch and validation correlation. Note: Data split information is not logged in this instance.

```text
Output:
2025-11-21 02:42:59 INFO     Training finished
2025-11-21 02:42:59 INFO     Find best model at epoch 29 with valid correation  0.743

```

--------------------------------

### Training Completion for CST3

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Reports the successful training of the CST3 model, including the best epoch, validation correlation, and confirmation of the training data split.

```text
Output:
2025-11-21 02:43:44 INFO     Training finished
2025-11-21 02:43:44 INFO     Find best model at epoch 8 with valid correation  0.903
2025-11-21 02:43:46 INFO     Training size is provided: 0.8
2025-11-21 02:43:46 INFO     Split the data for training and validation

```

--------------------------------

### Training Completion for GPCPD1

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Reports the successful training of the GPCPD1 model, including the best epoch, validation correlation, and confirmation of the training data split.

```text
Output:
2025-11-21 02:42:17 INFO     Training finished
2025-11-21 02:42:17 INFO     Find best model at epoch 24 with valid correation  0.956
2025-11-21 02:42:19 INFO     Training size is provided: 0.8
2025-11-21 02:42:19 INFO     Split the data for training and validation

```

--------------------------------

### Train Cell2Net Models for Genes

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Iterates through a list of genes to train Cell2Net models. It initializes the model, loads pretrained weights, trains for a specified number of epochs with early stopping, evaluates performance, and saves results. This snippet is designed for training individual gene models and includes evaluation and saving steps.

```python
# We only train Cell2net for one gene to save time
for gene in genes[:1]:    
    print('Training model for gene:', gene)

    cn.utils.set_random_seed(42)
    model = cn.pd.model.Cell2Net(mdata=mdata,
                                 gene=gene, 
                                 covariates=['total_counts_rna_log', 'total_counts_atac_log'])

    # load pretrained weights for the sequence encoder
    model.module.seq_encoder.load_state_dict(pretrained_state_dict)

    model.train(max_epochs=40, device_name='cuda:1', 
                batch_size=16, train_idx=train_idx, 
                valid_idx=valid_idx, 
                num_workers=4,
                verbose=False)

    model.save(dir_path=f"{out_dir}/model")

    # set the model with the best checkpoint
    model.module.load_state_dict(model.check_point)

    # Evaluate the model for training and validation dataset
    train_pred = model.predict(model.mdata[train_idx])
    train_true = model.mdata[train_idx]["rna"].layers["counts"].todense().A1

    valid_pred = model.predict(model.mdata[valid_idx])
    valid_true = model.mdata[valid_idx]["rna"].layers["counts"].todense().A1

    df_train = pd.DataFrame({'true': train_true, 'pred': train_pred, 'data': 'train'})
    df_valid = pd.DataFrame({'true': valid_true, 'pred': valid_pred, 'data': 'valid'})

    df_train['true'] = np.log1p(df_train['true'])
    df_valid['true'] = np.log1p(df_valid['true'])
    df_train['pred'] = np.log1p(df_train['pred'])
    df_valid['pred'] = np.log1p(df_valid['pred'])

    fig, axes = plt.subplots(1, 2, figsize=(8, 4))

    sns.scatterplot(data=df_train, x='true', y='pred', ax=axes[0], label='train')
    sns.scatterplot(data=df_valid, x='true', y='pred', ax=axes[1], label='valid')

    axes[0].set_xlabel("Observation (log1p)")
    axes[0].set_ylabel("Prediction (log1p)")
    axes[1].set_xlabel("Observation (log1p)")
    axes[1].set_ylabel("Prediction (log1p)")

    # save figure
    plt.savefig(f'{out_dir}/plot/{gene}.png', dpi=300)
    plt.close()

    np.savez(f'{out_dir}/prediction/{gene}.npz', 
             best_valid_corr=model.best_valid_corr,
             train_pred=train_pred,
             train_true=train_true,
             valid_pred=valid_pred,
             valid_true=valid_true)
```

--------------------------------

### Load Peak-to-Gene DataFrames

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Reads CSV files containing peak-to-gene links from Signac, SCENT, SCARlink, and cell2net into pandas DataFrames. Note that cell2net uses a different index column.

```python
# read peak-to-gene links from Signac, SCENT, SCARlink, and cell2net
df_signac = pd.read_csv(f"./signac.csv")
df_scent = pd.read_csv(f"./scent.csv")
df_scarlink = pd.read_csv(f"./scarlink.csv")
df_cell2net = pd.read_csv(f"./cell2net.csv", index_col=0)
```

--------------------------------

### Cell2Net Training Output

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

This snippet shows the typical output logs during Cell2Net model training, including information about data splitting, training progress, and the best validation correlation found.

```log
Training model for gene: ATRN

```

```log
2025-11-21 02:19:49 INFO     Training size is provided: 0.8
2025-11-21 02:19:49 INFO     Split the data for training and validation
2025-11-21 02:20:40 INFO     Training finished
2025-11-21 02:20:40 INFO     Find best model at epoch 37 with valid correation  0.812
2025-11-21 02:20:41 INFO     Training size is provided: 0.8
2025-11-21 02:20:41 INFO     Split the data for training and validation

```

```log
Training model for gene: PLCB1

```

```log
2025-11-21 02:21:20 INFO     Training finished
2025-11-21 02:21:20 INFO     Find best model at epoch 32 with valid correation  0.932
2025-11-21 02:21:21 INFO     Training size is provided: 0.8
2025-11-21 02:21:21 INFO     Split the data for training and validation

```

```log
Training model for gene: LAMP5

```

```log
2025-11-21 02:21:56 INFO     Training finished
2025-11-21 02:21:56 INFO     Find best model at epoch 25 with valid correation  0.441
2025-11-21 02:21:57 INFO     Training size is provided: 0.8
2025-11-21 02:21:57 INFO     Split the data for training and validation

```

```log
Training model for gene: AL050403.2

```

```log
2025-11-21 02:22:33 INFO     Training finished
2025-11-21 02:22:33 INFO     Find best model at epoch 36 with valid correation  0.693
2025-11-21 02:22:34 INFO     Training size is provided: 0.8
2025-11-21 02:22:34 INFO     Split the data for training and validation

```

```log
Training model for gene: ISM1

```

```log
2025-11-21 02:23:08 INFO     Training finished
2025-11-21 02:23:08 INFO     Find best model at epoch 12 with valid correation  0.382
2025-11-21 02:23:10 INFO     Training size is provided: 0.8
2025-11-21 02:23:10 INFO     Split the data for training and validation

```

```log
Training model for gene: MACROD2

```

```log
2025-11-21 02:23:46 INFO     Training finished
2025-11-21 02:23:46 INFO     Find best model at epoch 15 with valid correation  0.514
2025-11-21 02:23:47 INFO     Training size is provided: 0.8
2025-11-21 02:23:47 INFO     Split the data for training and validation

```

```log
Training model for gene: SLC24A3

```

```log
2025-11-21 02:24:23 INFO     Training finished
2025-11-21 02:24:23 INFO     Find best model at epoch 17 with valid correation  0.167
2025-11-21 02:24:25 INFO     Training size is provided: 0.8
2025-11-21 02:24:25 INFO     Split the data for training and validation

```

```log
Training model for gene: RIN2

```

--------------------------------

### Process Metacells for Visualization

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/02_prepare_data.ipynb

Applies normalization, PCA, neighbor graph construction, and UMAP embedding to metacells for visualization and quality assessment. Requires the scanpy library.

```python
sc.pp.normalize_total(mdata_bulk['rna'])
sc.tl.pca(mdata_bulk['rna'], n_comps=30, use_highly_variable=True)
sc.pp.neighbors(mdata_bulk['rna'])
sc.tl.umap(mdata_bulk['rna'])
```

--------------------------------

### Load Pretrained Sequence Encoder for Evaluation

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Loads the saved sequence encoder weights from a file. This is used to reload the best model for final evaluation and prediction visualization.

```python
# load best model weights
pretrained_state_dict = torch.load('./pretrained_seq2acc.pth')
model.module.seq_encoder.load_state_dict(pretrained_state_dict)
```

--------------------------------

### Examine MuData Object

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Displays the structure and summary of the loaded MuData object. This helps in understanding the dimensions and modalities of the dataset.

```python
mdata
```

--------------------------------

### Display SCARlink DataFrame Head

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Shows the first two rows of the SCARlink peak-to-gene prediction DataFrame to inspect its structure and content.

```python
df_scarlink.head(n=2)
```

--------------------------------

### Training Completion for MAFB

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Confirms the completion of training for MAFB, providing the best epoch, validation correlation, and details on the training data split.

```text
Output:
2025-11-21 02:46:08 INFO     Training finished
2025-11-21 02:46:08 INFO     Find best model at epoch 3 with valid correation  0.838
2025-11-21 02:46:10 INFO     Training size is provided: 0.8
2025-11-21 02:46:10 INFO     Split the data for training and validation

```

--------------------------------

### Training Completion for RNF24

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Confirms the completion of training for RNF24, providing the best epoch, validation correlation, and details on the training data split.

```text
Output:
2025-11-21 02:41:27 INFO     Training finished
2025-11-21 02:41:27 INFO     Find best model at epoch 14 with valid correation  0.873
2025-11-21 02:41:28 INFO     Training size is provided: 0.8
2025-11-21 02:41:28 INFO     Split the data for training and validation

```

--------------------------------

### Create Performance Comparison Plot

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/06_evaluate_p2g.ipynb

Generates a line and point plot comparing method performance. Requires the ggplot2 and cowplot libraries. Adjust plot dimensions using options().

```python
options(repr.plot.height = 4, repr.plot.width = 10)

p <- ggplot(data = df_final, aes(x = n_links, y = odds_ratio)) +
    geom_line(aes(color = method)) +
    geom_point(aes(color = method)) +
    facet_wrap(~data, scales = "free") +
    xlab("") + ylab("") +
    theme_cowplot() +
    theme(legend.title = element_blank())

print(p)
```

--------------------------------

### Cell2Net Training Output for AP000317.1 (Second Instance)

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/03_train_cell2net.ipynb

Shows the training completion, best model epoch, and validation correlation for gene AP000317.1. Indicates the training size and data split for validation.

```text
Output:
2025-11-21 02:59:51 INFO     Training finished
2025-11-21 02:59:51 INFO     Find best model at epoch 8 with valid correation  0.299
2025-11-21 02:59:53 INFO     Training size is provided: 0.8
2025-11-21 02:59:53 INFO     Split the data for training and validation

```

--------------------------------

### Load Data and Extract Target Genes

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Loads preprocessed multiome data from a specified file and extracts a unique list of genes that have valid peak-to-gene associations. These genes will serve as the targets for model training.

```python
mdata = md.read_h5mu(input_data)
genes = mdata.uns['peak_to_gene']['gene'].unique().tolist()
```

--------------------------------

### Create Train/Validation Split

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/03_train.ipynb

Splits the cell observations into training (80%) and validation (20%) sets using a fixed random state for reproducibility. This is crucial for evaluating model performance and preventing overfitting.

```python
train_idx, valid_idx = train_test_split(
    mdata.obs_names.values.tolist(),
    train_size=0.8,
    random_state=42)
```

--------------------------------

### Display cell2net DataFrame Head

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Shows the first five rows of the cell2net peak-to-gene prediction DataFrame to inspect its structure and content.

```python
df_cell2net.head(n=5)
```

--------------------------------

### RNA Data Preprocessing

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/02_prepare_data.ipynb

Applies total count normalization, log transformation, and PCA to RNA data. Ensure the AnnData object `mdata['rna']` is properly loaded.

```python
sc.pp.normalize_total(mdata['rna'])
sc.pp.log1p(mdata['rna'])
sc.tl.pca(mdata['rna'], n_comps=30, use_highly_variable=True)
```

--------------------------------

### Import Libraries for Cell2Net Correlation Analysis

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/pbmc/04_get_correlation.ipynb

Imports necessary Python libraries for data manipulation, numerical operations, statistical analysis, and plotting. Ensures warnings are ignored and sets Mudata options.

```python
import warnings
warnings.filterwarnings("ignore")
import os
import mudata as md
import pandas as pd
import numpy as np
import seaborn as sns
from scipy.stats import spearmanr
import matplotlib.pyplot as plt
md.set_options(pull_on_update=False)
```

--------------------------------

### Extract Essential Columns

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/05_process_p2g.ipynb

Selects 'Peak1', 'Peak2', and 'score' columns from Signac, Scent, ScarLink, and Cell2Net dataframes for evaluation.

```python
df_signac = df_signac[['Peak1', 'Peak2', 'score']]
df_scent = df_scent[['Peak1', 'Peak2', 'score']]
df_scarlink = df_scarlink[['Peak1', 'Peak2', 'score']]
df_cell2net = df_cell2net[['Peak1', 'Peak2', 'score']]
```

--------------------------------

### Train Sequence Encoder

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Initiates the training process for the sequence encoder model using specified training and validation dataframes and hyperparameters. Ensure df_train and df_valid are pre-loaded.

```python
model.train(df_train=df_train, df_valid=df_valid, max_epochs=100, 
            batch_size=512, weight_decay=1e-04, lr=3e-04)
```

--------------------------------

### Visualize ATAC UMAP

Source: https://github.com/pinellolab/cell2net/blob/main/docs/tutorials/k562/01_pretrain_seq_encoder.ipynb

Generates a UMAP visualization for ATAC data. Ensure 'scanpy' is imported as 'sc'.

```python
sc.pl.umap(mdata["atac"])
```