Quick Start

This guide summarizes the end-to-end workflow for reproducing the experiments described in the X-Pert manuscript. Each stage corresponds to a curated Jupyter notebook that you can launch after installing the dependencies.

Workflow Overview

1. Prepare the Environment

   1. Install scGPT and X-Pert as described in Installation Guide.

   2. Download the required perturbation datasets (links provided inside each notebook).

2. Construct Perturbation Datasets

   • Genetic perturbation preprocessing

   • Chemical perturbation preprocessing

   • Joint (co-train) perturbation preprocessing

3. Train Perturbation Models

   • Train X-Pert on genetic perturbations

   • Train X-Pert on chemical perturbations

4. Explore the Perturbverse

   • Visualize latent perturbation embeddings

Minimal Python Check

After completing the environment setup, verify that both scGPT and X-Pert import correctly:

import scgpt
import xpert

print(f"scGPT version: {scgpt.__version__}")
print(f"X-Pert version: {xpert.__version__}")

High-level Python APIs for perturbation modeling are actively evolving. For complete, reproducible pipelines—including data preparation, training, and visualization—use the tutorial notebooks linked above.

Next Steps

• Switch to the tutorial notebooks listed above for full data-to-results pipelines.

• Follow ongoing updates and discussions on GitHub: https://github.com/Chen-Li-17/X-Pert

• Reach out to the maintainers at chen-li21@mails.tsinghua.edu.cn if you run into issues.