pca.json

{
  "tool_id": "pca",
  "name": "Principal Component Analysis",
  "category": "dimensionality_reduction",
  "framework": "scanpy",
  "function": "sc.pp.pca",
  "valid_after": ["feature_selection"],
  "valid_before": ["integration", "neighbor_graph"],
  "paradigms": ["all"],
  "parameters": {
    "n_comps": {
      "type": "int",
      "default": 50,
      "range": [10, 100],
      "guidance": "Number of principal components to compute. 30-50 is standard. Use elbow plot of variance ratio (sc.pl.pca_variance_ratio) to determine how many PCs capture meaningful variance."
    },
    "use_highly_variable": {
      "type": "bool",
      "default": true,
      "guidance": "If true, restrict PCA to highly variable genes only. Should be true after HVG selection."
    },
    "svd_solver": {
      "type": "string",
      "default": "arpack",
      "options": ["arpack", "randomized", "lobpcg"],
      "guidance": "SVD solver. 'arpack' is default and reliable. 'randomized' is faster for large datasets (>50K cells)."
    }
  },
  "outputs": {
    "cell_embeddings": "adata.obsm['X_pca']",
    "variance_ratio": "adata.uns['pca']['variance_ratio']",
    "loadings": "adata.varm['PCs']"
  },
  "validation": {
    "pc1_max_variance": 0.5,
    "check": "First PC should explain <50% of variance. If it does, it likely reflects a technical artifact (e.g., total counts, cell cycle) rather than biology."
  },
  "provenance_captures": ["n_comps", "use_highly_variable", "svd_solver", "total_variance_explained"]
}