proteomics-de

Files (7)

• SKILL.md
• test_diann.tsv · examples/
• test_metadata.csv · examples/
• test_metadata_diann.csv · examples/
• test_proteinGroups.txt · examples/
• proteomics_de.py
• test_proteomics_de.py · tests/

SKILL.md (first lines)

# 🥚 Proteomics Differential Expression

This skill performs differential expression analysis on label-free quantitative (LFQ) intensity data from MaxQuant and DIA-NN outputs, including preprocessing, imputation, statistical testing, and visualization.

---

## Domain Decisions

### 1. Multi-format Input Support
- Supports **MaxQuant `proteinGroups.txt`**
  - Automatic filtering of reverse hits, contaminants, and site-only identifications
- Supports **DIA-NN output**
  - Automatically extracts protein IDs and `.raw` intensity columns

---

### 2. Preprocessing Strategy
- MaxQuant:
  - Filters:
    - `Reverse`
    - `Potential contaminant` / `Contaminant`
    - `Only identified by site`
- DIA-NN:
  - Extracts protein identifiers and intensity matrix directly

---

### 3. Intensity Transformation
- LFQ intensities are transformed using **log2 scaling**
- Ensures approximate normality for downstream statistical testing

---

### 4. Missing Value Imputation
- Uses **down-shifted Gaussian imputation**
  - Mean shifted by: `median - shift × std`
  - Default:
    - `shift = 1.8`
    - `scale = 0.3`
- Assumption:
  - Missing values represent **low-abundance proteins**

---

### 5. Statistical Testing
- Two-sample **t-test** between treatment and control groups
- Default degrees of freedom:
  - `df = 4` (for 3 vs 3 replicates)

---

### 6. s0-based FDR Correction
- Uses **s0-based thresholding** to stabilize variance
- Combines:
  - log2 fold change
  - p-value
- Based on:
  - Giai Gianetto et al. (2016)

---

### 7. Significance Thresholding
- Default:
  - `FDR = 0.05`
  - `s0 = 0.1`
- Produces:
  - Adjusted significance boundary (used in volcano plot)

---

### 8. Visualization Outputs
- PCA plot
- Volcano plot (with s0 curve)
- Imputation distribution comparison

---

## Safety Rules

- **Local-first**
  - No data upload without explicit user consent

- **Statistical caution**
  - Statistical results should be interpreted with caution and not overinterpreted
  - Avoid drawing conclusions beyond what the data supports

- **Missing data assumptions**
  - Imputation assumes missing values correspond to low abundance
  - May not hold in all experimental designs

- **Small sample limitations**
  - t-test reliability depends on sufficient replicates

- **Reproducibility**
  - All parameters and commands are logged

- **No hallucinated science**
  - All methods are based on established proteomics workflows

---

## Agent Boundary

### This skill DOES:
- Perform differential expression analysis on LFQ proteomics data
- Handle MaxQuant and DIA-NN outputs
- Generate statistical results and visualizations
- Produce reproducible reports

---

### This skill DOES NOT:
- Process raw mass spectrometry data (e.g. RAW files)
- Perform peptide identification or database search
- Conduct pathway or functional enrichment analysis
- Provide biological interpretation of results

---

From ClawBio/ClawBio, MIT.

More from ClawBio/ClawBio