Skip to content

Extensions

Three places hold what a view has no field for. Pick by who else should read the value.

Place Holds Readers
cv_params A string per row, named below Anyone can list it; the type is the reader's concern
x_<namespace>_<name> column A string column of your tool, typed by the reader Queryable like any column, and the validator accepts it
uns of an AnnData view, or a Parquet footer key Results about the whole file Your own tooling

A column named otherwise makes qpxc validate warn (undeclared_column), and a writer refuses or drops it with a warning.

cv_params names

An entry is {cv_name, cv_value}. Use the snake_case name of a source column or a PSI-MS term, and put a tool's private meaning behind its name: diann_quantity_quality. Scores & CV Terms lists the shared names, and Converter Coverage those each converter writes. A value used for filtering or joining belongs in a column, where the type and the statistics help.

Extension columns

x_ columns are declared to a writer, and are nullable strings:

with FeatureWriter(path, extra_columns=["x_mytool_cluster"]) as writer:
    writer.write_batch(records)

The sample view also takes any column: the SDRF's characteristics[...] and factor value[...] arrive there as organism_part and factor_<name>.

Targeted runs: PRM, SRM and heavy standards

QPX has no Skyline converter and no field for a spiked amount, and a heavy channel needs a sample, or the pg check fails. This convention validates:

  • Give the standard a channel and its own row in the sample view, sample_accession SIS-1, with x_ or factor_ columns for the amount and unit.
  • On the feature row, the heavy peptidoform's cv_params holds spiked_amount, spiked_amount_unit and paired_peptidoform, the endogenous peptide it stands for.
  • A transition's area is a row of the fragment view.

Curve fits: thermal profiling and dose response

The raw values per temperature or dose fit the views: one channel or run per point, with the point in factor_<name> of the sample view. A fit's result (Tm, EC50) describes a protein across the design, so it goes in the .de.h5ad or .ae.h5ad view:

adata.uns["fit_results"] = pd.DataFrame(
    {
        "protein": ["P12345"],
        "condition": ["treated"],
        "model": ["sigmoid"],
        "parameter": ["Tm"],
        "value": [52.3],
        "stderr": [0.4],
        "r2": [0.98],
        "n_points": [10],
    }
)

This is a recommendation, which qpxc validate does not read.