scDeepSim

Overview

This repository contains research code for the ongoing project about controllable diffusion models for single-cell data simulation.

A Python package scdeepsim (located under ./scdeepsim/), is built for convenient and reproducible research experiments. It provides a unified interface for latent diffusion modeling and controlled data simulation.

Notebooks under ./experiments/ show example analyses.

Installation

The package is defined by ./scdeepsim/pyproject.toml. To install from the repository root:

python -m pip install -U pip
python -m pip install -e "./scdeepsim"

Project layout

• scdeepsim/src/scdeepsim/: package source
  • diffusion_model.py: denoising backbone (MLP U-Net style) + classifier-free guidance logic
  • diffusion_core.py: diffusion schedules, losses, DDPM/DDIM sampling
  • lightning_diffusion.py: Lightning training loop + sampling convenience
  • dataset.py: ScDataset / ScDataModule
  • ae.py: autoencoder for single-cell data
  • transform.py: preprocessing scalers
  • plot.py: UMAP plotting helpers
  • control.py: controlled data simulation utilities
• experiments/: research notebooks

License & attribution

• This repository is released under the MIT License (see LICENSE).
• Upstream inspiration/adaptation: parts of the diffusion implementation were inspired by and/or adapted from lucidrains/denoising-diffusion-pytorch (MIT). See THIRD_PARTY_NOTICES.md for details.