Sdrf:brainstorm

SDRF Brainstorming Workflow

You are helping the user plan their SDRF annotation BEFORE creating the file. This is a collaborative thinking session, not a file generation task.

Step 1: Understand the Experiment

Ask the user about (if not already provided):

1. What organism? (human, mouse, rat, plant, microbiome, etc.)
2. What technology? (DDA, DIA, TMT, SILAC, label-free, Olink, SomaScan)
3. What is being compared? (disease vs control, treatment vs untreated, time course)
4. What tissue/sample type? (tissue biopsy, cell line, plasma, FFPE, etc.)
5. How many samples per group?
6. Is there fractionation? (high-pH RP, SAX, gel bands)
7. Is there a publication or PXD to reference?

Step 2: Recommend Templates

Use the sdrf:templates decision tree to select the right combination. Reference the 5 template layers:

1. Technology (required): ms-proteomics or affinity-proteomics
2. Organism (recommended): human, vertebrates, invertebrates, or plants
3. Experiment type (if applicable): dia-acquisition, cell-lines, single-cell, immunopeptidomics, crosslinking
4. Clinical/Domain (if applicable): clinical-metadata, oncology-metadata
5. Metaproteomics (special): metaproteomics + child (human-gut, soil, water)

Read spec/sdrf-proteomics/sdrf-templates/templates.yaml to confirm template names and current versions.

Present the recommendation:

Your experiment: [description]

Recommended templates:
  1. ms-proteomics (required — mass spectrometry experiment)
  2. human (organism is Homo sapiens)
  3. clinical-metadata (patient samples with treatment data)
  4. oncology-metadata (cancer study — adds tumor staging)

This combination requires these columns: [read from TERMS.tsv, filter by template names in usage]
And recommends these additional columns: [read from template YAMLs for recommended columns]

Read spec/sdrf-proteomics/TERMS.tsv and filter by the selected template names to list the columns.

Step 3: Search for Similar Experiments

Find reference datasets to learn from:

Search PRIDE for similar experiments:
  mcp PRIDE → search_extensive(query="")

Search publications for standard experimental designs:
  mcp PubMed → search_articles(query=" AND proteomics")

Search bioRxiv for recent preprints:
  mcp bioRxiv → search_preprints(category="biochemistry" or "cell biology", recent_days=180)

Present findings:

Similar datasets found:
  - PXD012345: TMT phosphoproteomics of breast cancer (24 samples, 12 fractions)
  - PXD023456: Label-free DIA of liver cancer tissue (30 patients)

Common design patterns in this field:
  - Typical sample size: 10-30 per group
  - Common labels: TMT, label-free DIA, SILAC
  - Standard fractionation: 12-24 high-pH RP fractions
  - Most include: age, sex, disease staging

Step 4: Recommend Metadata Columns

Present a complete column plan organized by importance:

Must Have (required by templates)

Columns required by the selected template combination. For each: explain what it is, what ontology to use, and give examples.

Should Have (recommended for this experiment type)

Columns that 70%+ of similar experiments include. For each: explain why it adds value.

Nice to Have (optional but valuable)

Columns that would increase reusability and findability. For each: explain the benefit.

Factor Values

Discuss what the experimental comparison is:

• What variable is being tested?
• Are there multiple factors? (e.g., disease × treatment)
• What will the statistical comparison be?

Step 5: Discuss Design Considerations

Raise potential issues proactively:

Batch Effects

• Will all conditions be processed together or separately?
• Are label channels balanced across conditions?
• Is the instrument assignment confounded with the experimental variable?

Replication

• How many biological replicates per condition?
• Are technical replicates needed?
• Is the sample size adequate for statistical testing?

Labeling Strategy (if TMT/iTRAQ)

• How will samples be assigned to channels?
• Will there be a pooled reference channel?
• How many TMT sets are needed?

Fractionation

• How many fractions per sample?
• What method? (high-pH RP, SAX, SCX, gel)
• How does this affect the total number of SDRF rows?

Step 6: Calculate SDRF Dimensions

Help the user understand the scale:

Your SDRF will have:
  Rows: [samples] × [fractions] × [label channels] × [technical replicates]

  Example: 20 patients × 12 fractions × 1 (label-free) × 1 replicate = 240 rows
  Example: 20 patients × 12 fractions × 10 (TMT10plex) × 1 replicate = 2,400 rows

  Columns: ~15 required + ~8 recommended + factor values = ~25 columns

Step 7: Summarize the Plan

Create a clear annotation plan the user can follow:

## SDRF Annotation Plan for [experiment]

Templates: ms-proteomics + human + oncology-metadata
Rows: ~240 (20 patients × 12 fractions)
Columns: 26

Required metadata to collect:
  - Patient demographics: age, sex (from clinical records)
  - Diagnosis: specific cancer subtype (from pathology)
  - Tumor staging: TNM stage, grade (from clinical records)
  - Tissue type: primary tumor vs adjacent normal

Technical metadata (from instrument):
  - Instrument model, fragmentation method
  - Mass tolerances, collision energy
  - Label type and channel assignments

Factor values: disease (tumor vs normal)

Next step: Run /sdrf:annotate to create the file

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: bigbio
• Source: bigbio/sdrf-skills
• License: MIT
• Homepage: https://sdrf.quantms.org

Install and usage instructions live in the source repository linked above.