usage.md

Run scGenAI

This document explains how to run scGenAI for training, fine-tuning, and prediction tasks. You can use scGenAI in two ways:

1. Running through Python scripts: Leverage the provided tutorials in the tutorials/ folder.
2. Running through the command line: Execute tasks directly via command-line commands.

Option 1: Running scGenAI through Python

You can run scGenAI through the provided tutorial notebooks or by writing your own Python script. The tutorials cover common workflows, such as training, fine-tuning, and prediction.

Tutorials Jupyter Notebook

As a quick start, we strongly suggest to begin with the following Jupyter notebooks in the tutorials/ folder guide you through various workflows:

• Training and Prediction with Biofunction Context Method: TrainData_biofounction_context.ipynb
• Training and Prediction with Genomic Context Method: TrainData_genomic_context.ipynb
• Training and Prediction with Random Context Method: TrainData_random_context.ipynb
• Training and Prediction using Multi-omics Data: TrainData_MultiOmics.ipynb
• Fine-tuning a pre-trained model: FinetuneData_random_context.ipynb

To explore these, simply open the Jupyter notebook of your choice and follow the steps to run the training, prediction, or fine-tuning process.

Python Examples

Here are a few examples of how to run scGenAI using Python:

Training

from scGenAI.training.train import run_training_from_config

# Path to your configuration file
config_path = "/packagdir/config_templates/config_Train_genelist_template_llama.yaml"

# Run training
run_training_from_config(config_path)

Prediction

from scGenAI.prediction.predict import run_prediction_from_config

# Path to your configuration file
config_path = "/packagdir/config_templates/config_random_context_template_llama.yaml"

# Run training
run_prediction_from_config(config_path)

Fine-tune

from scGenAI.finetuning.finetune import run_finetune_from_config

# Path to your configuration file
config_path = "/packagdir/config_templates/config_Finetune_random_context_template_llama.yaml"

# Run training
run_finetune_from_config(config_path)

Login Huggingface (ONLY required for the FIRST TIME use of llama model template)

Logging into Hugging Face with a token is required when using the LLaMA model for the first time. This requirement is specific to LLaMA and does not apply to other model architectures supported by scGenAI.

from huggingface_hub import login
import getpass

def login_to_huggingface():
    """
    Prompts the user to securely enter their Hugging Face token and logs them into the Hugging Face Hub.
    """
    token = getpass.getpass(prompt='Please enter your Hugging Face token: ')
    login(token=token)
    print("Successfully logged in to Hugging Face!")

# Call the function to log in
login_to_huggingface()

• For detailed information on input files and parameters, see the Input Documentation.
• To learn more about how to configure scGenAI, visit the Configuration Guide.

---

Option 2: Running scGenAI from the Command Line

You can also run scGenAI directly from the command line by providing arguments for different modes (Train, Predict, Finetune).

The CLI supports the following commands:

• Train a model:

  scgenai train --config_file <path_to_config.yaml>

• Make predictions:

  scgenai predict --config_file <path_to_config.yaml>

• Fine-tune a pre-trained model:

  scgenai finetune --config_file <path_to_config.yaml>

---

Next Step: Output Files

Once the running is finished, the output model and files can be found in the path defined in the config file. Please read the doc of Output Files for details