dynamo-preprocess

Dynamo Preprocess

Goal

Preprocess an AnnData object with the current dynamo.preprocessing.Preprocessor API, choose the correct recipe branch for the downstream task, customize kwargs before execution when needed, and validate that the expected keys and embeddings exist for later dynamo analysis.

Quick Workflow

1. Inspect the user's data, environment, and downstream goal.
2. Choose the recipe branch that matches the goal instead of defaulting blindly to the notebook's first example.
3. Use Preprocessor.preprocess_adata(...) for the common case.
4. If the user needs customization, call config_*_recipe(...), mutate kwargs, then run the matching recipe-specific method.
5. If the user needs debugging or a custom pipeline, run the preprocessing steps individually in notebook order.
6. Validate obs, var, layers, and obsm keys before treating preprocessing as complete.

Interface Summary

• Preprocessor.preprocess_adata(adata, recipe='monocle', tkey=None, experiment_type=None) is the main wrapper.
• The live source dispatches recipe across five branches: monocle, seurat, sctransform, pearson_residuals, monocle_pearson_residuals.
• config_monocle_recipe(adata, n_top_genes=2000) is the only recipe config in this notebook family that exposes an extra tuning argument directly in the signature.
• The constructor exposes many overridable callables and kwargs dictionaries, but the notebook mainly mutates: filter_cells_by_outliers_kwargs, filter_genes_by_outliers_kwargs, and select_genes_kwargs.

Read references/source-grounding.md before documenting parameters in more detail or when you need the exact inspected signatures.

Recipe Selection

• Use monocle as the default when the goal is standard dynamo preprocessing for velocity or vector-field analysis.
• Use seurat when the user specifically wants Seurat-style highly variable gene selection inside the current Preprocessor wrapper.
• Use sctransform only when the user explicitly wants that transformation and the environment has KDEpy.
• Use pearson_residuals when the goal is HVG selection and PCA on adata.X, not layer-preserving velocity normalization.
• Use monocle_pearson_residuals when the user wants Pearson-residual-based feature selection and PCA but still needs monocle-style normalized layers for downstream velocity analysis.