Single Cell Manifold Generator (SCMG)

SCMG is a suite of deep learning models designed to interpret, generate, and predict the molecular basis of cell states and their transitions.

Key Features

• 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.

Contents

• Installation
• Trained model parameters
• Reference and tutorial datasets
• Tutorials
  • Contents
    • Visualize the global cell state manifold
      • Visualize cell type distributions
      • Visualize gene expression patterns
    • Zero-shot integration of scRNA-seq datasets
    • Project single-cell states onto the global cell state manifold
      • Project a dataset of natural cell states
      • Project a dataset of transdifferentiation
    • Conditioned generation of cell states
      • Generation of cell types
      • Generation of continuous cell state trajectories
    • Predict causal genes for cell state transitions
    • Universal Decomposition of Perturbation Effects
      • Load and preprocess the perturb-seq dataset
      • Project perturbation effects onto the principal axes of cell state transitions
      • Grouping of perturbations into universal clusters
    • Few-shot Prediction of Perturbation Effects
      • Learn a universal embedding of perturbations
      • Few shot prediction
• API
  • scmg
    • scmg package
      • Subpackages
        • scmg.model package
          • Submodules
            • scmg.model.basic module
              • MLP
                • MLP.forward()
            • scmg.model.causal_prediction module
              • CausalGenePredictor
                • CausalGenePredictor.calc_causal_scores()
            • scmg.model.cell_type_search module
              • CellTypeSearcher
                • CellTypeSearcher.search_ref_cell()
                • CellTypeSearcher.search_ref_cell_types()
            • scmg.model.contrastive_embedding module
            • scmg.model.manifold_generation module
              • sinusoidal_embedding()
              • ConditionalDiffusionModel
                • ConditionalDiffusionModel.forward()
                • ConditionalDiffusionModel.backward()
                • ConditionalDiffusionModel.generate()
              • RecurrentBlock
                • RecurrentBlock.forward()
              • MLPDenoiser
                • MLPDenoiser.forward()
              • train_diffusion_model()
              • generate_cells()
              • generate_transition_cells()
          • Module contents
        • scmg.preprocessing package
          • Submodules
            • scmg.preprocessing.data_standardization module
          • Module contents
      • Module contents