SCMG is a suite of deep learning models designed to interpret, generate, and predict the molecular basis of cell states and their transitions.
-
Global Manifold Construction
Build a well-integrated reference manifold of single-cell transcriptional states that captures cell-state relationships and gene expression patterns. The global gene expression patterns can be visualized here. -
Zero-Shot Dataset Integration
Integrate new scRNA-seq datasets without the need for model retraining. -
Zero-Shot Cell Projection
Project single-cells onto the global manifold for downstream analysis and comparison. -
Cell State Trajectory Generation
Generate continuous trajectories to model transitions between cell states. -
Causal Gene Prediction
Identify candidate causal genes driving transitions between specific cell states. -
Universal Decomposition of Perturbation Effects
Decompose perturbation effects into universal principal axes of cell state transition and perturbation classes. -
Few-shot Prediction of Perturbation Effects
Predict perturbation-induced cell state transition by few-shot learning.
The SCMG package can be installed from pip. The installation takes one to a few minutes. The detailed instructions for installation are available here
This package is compatible with major operating systems that support PyTorch, including Linux, macOS, and Windows.
The package dependencies are specified in pyproject.toml
The current version of the tested release is scmg1.0
Tutorials for the main functions of SCMG are available here. The running time for individual tutorials ranges from a few minutes (with GPU) to within one hour (with CPU only).
The scripts to reproduce the results reported in the manuscript are available here.
Global cell-state and gene-program representations reveal conserved and context-specific perturbation responses of cells. preprint.