CycSeq: Leveraging Cyclic Data Generation for Accurate Perturbation Prediction in Single-Cell RNA-Seq

Accepted at IJCAI 2025

---

Abstract

Understanding and predicting the effects of cellular perturbations using single-cell technology is a critical problem and remains challenging in biotechnology. In this work, we leverage the concept of cyclic data generation and the recent advancements in neural architectures to introduce CycSeq, a deep learning framework capable of predicting single-cell responses under specified perturbation conditions across multiple cell lines and generating corresponding single-cell expression profiles. CycSeq addresses the challenge of learning heterogeneous perturbation responses from unpaired single-cell level gene expression observation by generating pseudo-pairs through cyclic data generation. Experimental results show that CycSeq could outperform current methods in perturbation prediction tasks evaluated by computational metrics($R^2$, MAE, etc.). CycSeq also tackles the issue of applying deep learning models to long-tail cell lines by utilizing a unified architecture that integrates information from multiple cell lines, enabling robust predictions even for cell lines with limited training data.

---

Introduction

CycSeq is a novel deep learning framework based on cycle-consistent generative adversarial networks (CycleGAN) for the integration and translation of single-cell perturbation datasets. This repository provides the official implementation of the methods described in our IJCAI 2025 accepted paper.

---

Paper

Title: Cycle-Consistent Generative Adversarial Networks for Single-Cell Perturbation Data Integration
Conference: International Joint Conference on Artificial Intelligence (IJCAI 2025)
Status: Accepted
Paper Link: (to be updated)

---

Installation

1. Clone the repository

git clone https://github.com/yczju/cycseq.git
cd cycseq

2. Create and activate a Python environment

We recommend using conda or venv for environment management.

conda create -n cycseq python=3.9
conda activate cycseq

3. Install dependencies

pip install -r requirements.txt

Key dependencies:

• torch >= 1.10
• numpy
• pandas
• tqdm
• matplotlib
• scanpy
• scipy
• tensorboard
• pycombat, inmoose

For GPU support, please ensure you have installed the correct CUDA version and compatible PyTorch.

---

Data Preparation

Please refer to the script/data_preparation.sh for detailed instructions on preparing training data.
The project supports both .csv and .npy formats for single-cell perturbation expression matrices.

---

Usage

Training

python code/CycleGAN4Seq/ijcai.py --num_res_blocks 2 --use_attention True --use_embedding False --use_batch_norm False --use_dropout True --name cycseq

Main Arguments

• --num_res_blocks: Number of residual blocks in encoder/decoder.
• --use_attention: Whether to use self-attention.
• --use_embedding: Whether to use embedding layers.
• --use_batch_norm: Whether to use batch normalization.
• --use_dropout: Whether to use dropout.
• --name: Model name for output files.

---

Results

• The model has been evaluated on multiple public single-cell perturbation datasets. Please refer to the paper and the output_model/ directory for detailed results.
• Training logs and loss curves can be visualized using TensorBoard (logs/ directory).

---

Citation

If you find this repository useful for your research, please cite our paper:

(to be updated)

---

Contact

For questions or collaboration, please contact:

• yichengliu@zju.edu.cn

---

Acknowledgements

This project is inspired by CycleGAN and related single-cell analysis tools.