biotender-max / affinity-proteomics

🧪 Affinity Proteomics Pipeline

You are Affinity Proteomics, a specialised ClawBio agent for Olink and SomaLogic SomaScan data analysis. Your role is to run platform-aware QC, differential abundance testing, and visualisation from affinity-based proteomics data.

Why This Exists

• Without it: Researchers must write bespoke scripts for each platform — Olink NPX and SomaLogic ADAT have completely different file formats, normalisation methods, and QC conventions
• With it: A single command handles both platforms with correct QC, normalisation, and analysis under a unified interface
• Why ClawBio: The existing proteomics-de skill handles mass-spectrometry LFQ data (MaxQuant/DIA-NN) and does not cover affinity-based platforms. This skill fills that gap

Core Capabilities

1. Dual-platform support: Olink NPX (CSV/Parquet) and SomaLogic ADAT under one interface
2. Platform-specific QC: Olink (QC_Warning, LOD, sample median) / SomaLogic (RowCheck, ColCheck, normalisation scale factors, MAD outlier filtering)
3. Differential abundance: t-test or Mann-Whitney U with Benjamini-Hochberg FDR correction
4. Visualisation: Volcano plot, heatmap (top N proteins), PCA plot
5. Structured reporting: Markdown report, result.json, per-protein TSV, reproducibility bundle
6. Skill Action Menu: result.json includes a workflow state plus read-only follow-up actions for compact report cards

Input Formats

CLI Reference

# Olink demo
python skills/affinity-proteomics/affinity_proteomics.py \
  --demo --platform olink --output /tmp/olink_demo

# SomaLogic demo
python skills/affinity-proteomics/affinity_proteomics.py \
  --demo --platform somascan --output /tmp/soma_demo

# Real Olink data
python skills/affinity-proteomics/affinity_proteomics.py \
  --platform olink --input data.csv --meta samples.csv \
  --group-col Group --contrast "Case,Control" --output results/

# Via ClawBio runner
python clawbio.py run affprot --demo --platform olink

Demo

python clawbio.py run affprot --demo --platform olink

Expected output: Differential abundance report for 80 samples (40 Case / 40 Control) across 40 proteins, with 5 truly differentially expressed proteins recovered, volcano plot, heatmap, PCA, and reproducibility bundle.

Output Structure

• report.md — markdown report with QC, differential abundance, and top-protein sections
• result.json — structured summary with chat_summary_lines, preferred_artifacts, workflow_state, and suggested_actions
• tables/diff_abundance.tsv — per-protein differential abundance table
• figures/volcano.png, figures/heatmap.png, figures/pca.png — standard demo figures
• reproducibility/ — command and software-version metadata

Suggested Actions

The demo result emits workflow_state.lifecycle: "ready" and offers two read-only actions: Top Proteins and Volcano Summary. In chat, the user sees those labels as numbered options; selecting one runs the stored structured request.

state_id is derived as a SHA-256 hash over a compact deterministic state payload: platform, contrast, protein counts, significant-protein direction counts, and the top protein rows carried in each action request. If a stored request's state_id no longer matches that payload, the skill returns a structured expired result instead of rendering a stale follow-up.

{
  "workflow_state": {
    "state_schema": "affinity_proteomics.workflow_state.v1",
    "state_id": "sha256:...",
    "lifecycle": "ready",
    "state_label": "differential-abundance-ready",
    "description": "OLINK differential abundance results for Case vs Control are available."
  },
  "suggested_actions": [
    {
      "action_id": "show-top-proteins",
      "label": "Top Proteins",
      "estimate": "~5s",
      "request": {
        "schema": "affinity_proteomics.action_request.v1",
        "action": "top-proteins",
        "state_schema": "affinity_proteomics.workflow_state.v1",
        "state_id": "sha256:...",
        "n": 5,
        "platform": "olink",
        "contrast": ["Case", "Control"],
        "total_proteins_tested": 40,
        "significant_proteins": 5,
        "proteins": [
          {"protein_id": "OID00001", "gene": "GENE1", "log2fc": 0.0, "padj": "0.00e+00"}
        ]
      }
    }
  ]
}

Dependencies

Required:

• somadata >= 1.2 — SomaLogic ADAT parsing
• scipy >= 1.10 — statistical tests
• statsmodels >= 0.14 — multiple testing correction
• matplotlib >= 3.7 — plotting
• seaborn >= 0.13 — heatmaps
• numpy >= 1.24 — numerical operations
• pandas >= 2.0 — data manipulation
• scikit-learn >= 1.3 — PCA dimensionality reduction for sample-level QC plots

Safety

• Local-first: All computation runs locally; no data uploaded
• Disclaimer: Every report includes the ClawBio medical disclaimer
• Platform-aware: Applies correct QC and normalisation per platform
• No hallucinated science: All thresholds trace to platform vendor documentation

Integration with Bio Orchestrator

Trigger conditions — the orchestrator routes here when:

• User mentions Olink, SomaLogic, SomaScan, NPX, ADAT, or affinity proteomics
• User provides an Olink NPX CSV or SomaLogic ADAT file

Chaining partners:

• proteomics-de: Complementary — handles mass-spec LFQ; this skill handles affinity platforms
• diff-visualizer: Downstream — enhanced visualisation of differential abundance results

Citations

• Assarsson et al. (2014) — Olink PEA technology
• Gold et al. (2010) — SOMAmer aptamer technology
• OlinkAnalyze — Official Olink R toolkit
• somadata — Python ADAT parser