Start here

Use this order: import and review sample metadata → inspect missingness → explicitly impute if needed → normalize/transform/scale → explore → test hypotheses or evaluate a predictive model. Save a derived dataset when you want later pages to use the transformed matrix.

Run applies the current controls. Changing an input does not update an existing plot until you run again. Cached view/style actions do not refit analysis. The optional Code for this plot panel records the applied inputs and settings, not pending edits.

Name in Python	Meaning
raw_data_df	The matrix supplied to the function: lipid IDs on rows, sample IDs on columns. It may already be processed; the name does not request automatic preprocessing.
sample_metadata_df	One row per sample, identified by sample_id; contains groups, subjects, batch or injection order.
lipid_metadata_df	Lipid annotations identified by lipid_id, such as lipid_class.
pca_params_dct, biomarker_params_dct	Dictionaries of named settings for the corresponding plotting workflow.
selected_lipid_ids_lst, lipid_ids_lst	Lists of lipid identifiers, not row positions.
lipid_sets_dct	Set names mapped to lists of lipid IDs, for enrichment.
color_profile	A lipidjaguar.plotting.ColorProfile containing colors, symbols and rendering settings.

_df, _dct, and _lst mean DataFrame, dictionary, and list. Names ending in _params without _dct refer to typed configuration objects. Pydantic validates configuration types and rejects unknown fields. Python uses True, False, None; REST JSON uses true, false, null. Historical keywords such as matrix, samples, config, and selected_ids remain aliases, but do not supply both names together.

Most metadata APIs expect a sample_id column. The differential-analysis function specifically takes sample_metadata_df.set_index("sample_id"). Never align groups merely by the visible row order. Scikit-learn itself usually expects samples × features; LipidJaguar performs the necessary transpose internally.