Skip to content

A result named by catalog keys

as_mudata() turns the MuData or AnnData APB2 wrote into a new MuData that holds only the fields the named catalog sets describe, each under its key. Any set works. write_h5mu writes the new object like any MuData, and the source object is left unchanged.

import mudata
from apb_catalog.api_pandas import DataCatalog

mdata = mudata.read_h5mu("diann.h5mu")        # DIA-NN 1.8.1: precursors and protein groups
keyed = DataCatalog(mdata, "abundance_alpha", "confidence_alpha", "apb_alpha").as_mudata()
keyed.write_h5mu("diann_keyed.h5mu")
keyed
MuData object with n_obs × n_vars = 4 × 12
  uns:  'catalog'
  2 modalities
    peptidoform_ion:    4 × 6
      var:  'q_value', 'library_q_value', 'peptide', 'peptidoform', 'peptidoform_ion', 'decoy', 'contaminant', 'protein_assignment', 'fasta_accessions'
      uns:  'catalog'
      layers:   'intensity', 'intensity_norm_settings', 'q_value', 'pep'
    protein:    4 × 6
      var:  'library_q_value', 'q_value', 'decoy', 'contaminant', 'protein_assignment', 'fasta_accessions'
      uns:  'catalog'
      layers:   'intensity', 'intensity_norm_settings[aggregation=top_n]', 'intensity_norm_settings[aggregation=maxlfq]', 'q_value'

What the new object holds

  • Sets: the named ones alone; apb_alpha brings the identifiers and the decoy and contaminant flags only when named
  • Modalities: one per level that has a cataloged field; levels["protein"].as_anndata() converts a single level
  • layers, var, obs: one entry per cataloged field, named by its key
  • Run and feature names: the result's own
  • X: empty, since no set says which layer is primary
  • A field two keys name: written under both; DIA-NN's Protein_Group is both protein_assignment and fasta_accessions
  • A field in varm: becomes a var column

Layers only

The abundance set's keys are all layers, so on its own it gives a MuData of layers; the run and feature names stay as obs_names and var_names:

DataCatalog(mdata, "abundance_alpha").as_mudata()
MuData object with n_obs × n_vars = 4 × 12
  uns:  'catalog'
  2 modalities
    peptidoform_ion:    4 × 6
      uns:  'catalog'
      layers:   'intensity', 'intensity_norm_settings'
    protein:    4 × 6
      uns:  'catalog'
      layers:   'intensity', 'intensity_norm_settings[aggregation=top_n]', 'intensity_norm_settings[aggregation=maxlfq]'

One key, two fields

DIA-NN 1.8 reports two protein intensities normalized as the run was set up: PG_Normalised, from the top N precursors, and PG_MaxLFQ. Both are intensity_norm_settings. The qualifier that tells them apart is added to the key, paired with its value:

Column DIA-NN's name
intensity PG_Quantity
intensity_norm_settings[aggregation=top_n] PG_Normalised
intensity_norm_settings[aggregation=maxlfq] PG_MaxLFQ
q_value PG_Q_Value

Two fields that no qualifier tells apart get their own name as well. With an SDRF and obs_annotation_alpha named, every factor column is group, so the factors treatment and dose become group[field=factor_value_treatment] and group[field=factor_value_dose].

The record in uns["catalog"]

Each column records where it came from:

protein = keyed["protein"]
protein.uns["catalog"]["layers"]["intensity_norm_settings[aggregation=maxlfq]"]
{'set': 'abundance_alpha', 'key': 'intensity_norm_settings', 'qualifiers': {'normalization': 'run_setting', 'aggregation': 'maxlfq', 'ms_level': 'ms2'}, 'location': 'layers', 'name': 'PG_MaxLFQ'}
  • layers, var, obs: for each column, its set, key, qualifiers, and the level's own name for the field and where it sat
  • absent: the required and recommended keys the level lacks. Spectronaut 21's ion level reports no unnormalized intensity, so with abundance_alpha the record reads {'layers': {'intensity': 'recommended'}}
  • source: the level's APB2 record, uns["apb"] as the result held it
  • MuData root: its own uns["catalog"] names the sets and holds the result's root record

From a Parquet or DuckDB result

as_mudata starts from the AnnData or MuData APB2 wrote. Write a result in another format to h5mu first:

from pathlib import Path
from apb2.api import read_parsed_levels, write_parsed_levels

write_parsed_levels(read_parsed_levels(Path("diann.parquet")), Path("diann.h5mu"))