Quality control of plasma samples by MS or Olink with the AnnData Proteomics Bridge (APB)

[wolski]

Title

Quality control of biological plasma samples and their MS or Olink measurements with APB

Abstract

APB2, the current implementation of the AnnData Proteomics Bridge, is a rules-driven framework for converting outputs from proteomics software into structured datasets. It grew out of ProteoBench's need to compare results from several quantification tools.

What APB does today:

• Preserves tool-specific columns
• Links protein, peptide and ion evidence, fragments for DIA
• Stores AnnData/MuData, Parquet or DuckDB
• Joins datasets and data levels in multimodal containers
• Enriches MuData; for example, apb fasta adds annotations, apb proteobench adds statistics and scores, and apb aggregate rolls up data from levels such as ions to proteins, creating new levels

Participants will extend APB to assess studies of hundreds of biological plasma samples. Quality signals in scope:

• Pre-analytical and biological sample quality
• Cellular contamination
• Expected-protein coverage
• Marker precision
• Abundance dynamic range
• Batch-associated shifts
• Agreement with reference datasets

The starting point is documented panels, for example Geyer et al. (2019), who used counted dilution series to identify erythrocyte and platelet contamination markers. With the Olink/proteoform project, #11, participants will test protein-level measures on a second modality and separate shared from platform-specific measures. Prepared APB datasets can be read from Python, R, Julia, or JavaScript.

Project Plan

Before the event, the organizers will:

• Quantify label-free DDA and DIA plasma datasets with FragPipe and DIA-NN
• Convert the resulting tables into MuData with APB2
• Prepare a small fixture, reference MuData datasets, and plasma panels
• Prepare mapped Olink datasets for cross-modality analysis with the Olink/proteoform project, #11

During the four project days, participants will:

1. define each measure's inputs, output, interpretation, data level, and validation dataset;
2. implement panel coverage, marker precision, and reference-dataset comparison;
3. define candidate cellular-contamination scores from documented panels, and implement at least one;
4. develop apb-plasma to annotate, compare, and score new measurements using reference MuData datasets and panels;
5. store the resulting annotations, scores, statistics, and relationships in an enriched MuData file;
6. validate protein-level quality measures that apply to both MS and Olink on at least two MS datasets and one Olink dataset;
7. record which measures need a platform-specific interpretation;
8. develop a quality-control visualization tool using Vitessce, Plotly, or another JavaScript library.

Contributions can be plasma datasets, panels, measure definitions, implementations, or visualizations.

During and after the hackathon:

• Document the measure definitions and validation results
• Present the workflow and findings to the EuBIC-MS community
• Publish software, panels, validation outputs and MuData examples in apb-plasma
• Prepare a manuscript on the cross-dataset and cross-modality evaluation

Technical Details

APB2 is written in Python and stores results as AnnData/MuData, Parquet, or DuckDB, so participants can work from Python, R, Julia, or JavaScript.

Alignment with #16, From Spreadsheet to Schema, which will implement MIAPE-AnnData and the Quantification CV:

• Make the objects that apb-plasma writes MIAPE-conformant
• Record transformations as Quant CV terms in uns
• Carry SDRF sample metadata into obs
• Validate the objects against the team's LinkML schema
• Use APB as a test implementation while the schema develops

The schema and validator remain independent of APB. APB's existing converters allow the MIAPE-AnnData representation to be tested with FragPipe, DIA-NN and other quantitative formats used by ProteoBench.

What the apb-plasma tool reads and writes:

• Three panel types: plasma biomarkers, cellular contamination markers, expected plasma proteins
• Reference MuData datasets linking protein, peptide and ion levels, fragments for DIA
• Enriched MuData exposing layers, var, varm, obs, obsm, obsp, and uns

What a measure has to satisfy:

• Participants define it; no formula is fixed in advance
• It uses only the matrices and annotations it needs
• It assumes no common raw scale for MS and Olink
• A cross-modality measure states a documented normalization, or a scale-free comparison such as rank concordance

The organizers will provide an in-browser MuData explorer. Participants will extend it into a plasma-specific view of the file apb-plasma generates, using Vitessce, Plotly, or other JavaScript libraries.

flowchart LR
    NEW["New biological plasma measurement<br/>MS MuData or mapped Olink table"]
    REF["Reference MuData datasets<br/>protein · peptide · ion<br/>fragment for DIA"]
    PANELS["Plasma panels in Parquet<br/>biomarkers · cellular contaminants · expected proteins"]
    APB["Python apb-plasma tool<br/>annotate · compare · score"]
    OUT["Enriched MuData<br/>annotations · scores · statistics · relationships"]
    VIEW["JavaScript viewer"]

    NEW --> APB
    REF --> APB
    PANELS --> APB
    APB --> OUT --> VIEW

Loading

Candidate datasets are public deposits. Accessions, designs, validation roles, and availability are listed in HACKATHON.md section 6, with the panels, measure definitions, and four-day plan.

Contact information

Witold E. Wolski — Functional Genomics Center Zurich (FGCZ), ETH Zurich / University of Zurich;
SIB Swiss Institute of Bioinformatics — witold.wolski@fgcz.uzh.ch

Sam van Puyenbroeck — CompOmics, VIB-UGent Center for Medical Biotechnology; Department of
Biomolecular Medicine, Ghent University — sam.vanpuyenbroeck@ugent.be

Tobias Kockmann — Functional Genomics Center Zurich (FGCZ), ETH Zurich / University of Zurich —
tobias.kockmann@fgcz.ethz.ch

Christoph B. Messner — Precision Proteomics Center, Swiss Institute of Allergy and Asthma Research (SIAF), University of Zurich, Davos, Switzerland — christoph.messner@siaf.uzh.ch