Tutorials

An cascade-reg workflow can be divided into the following two stages. We provide separate tutorials for each stage.

• Stage 1: Data preprocessing
  • Read data
  • Data format requirements
  • Cell and gene selection
  • Encode intervention regime
  • Encode technical covariates
  • Data normalization
  • Neighbor-based imputation
  • Configure dataset
  • Construct scaffold graph
  • Prepare gene function embeddings
  • Save processed data files
  • Afterwords
• Stage 2: Model training
  • Read preprocessed data
  • Build the CASCADE model
  • Run causal discovery
  • Remove remaining cycles
  • Model tuning
• Stage 3: Counterfactual prediction
  • Read data and model
  • Specify counterfactual condition
  • Run counterfactual prediction
  • Counterfactual differential expression comparison
  • Explain counterfactual prediction
• Stage 4: Intervention design
  • Prepare source data
  • Prepare target data
  • Define gene weights
  • Specify candidate genes
  • Run intervention design
  • Determine design score cutoff
  • Verify design with counterfactual prediction