⚖️ Differential Protein Abundance

Statistical testing for differentially abundant proteins between experimental conditions.

Core Capabilities

1. MSstats: Feature-level mixed models with group comparison
2. limma: Empirical Bayes moderated t-tests on protein-level data
3. proDA: Probabilistic handling of missing values
4. scipy/statsmodels (Python): T-test with FDR correction
5. Visualization: Volcano plots, heatmaps

CLI Reference

python omicsclaw.py run differential-abundance --demo
python omicsclaw.py run differential-abundance --input <protein_matrix.csv> --output <dir>

Algorithm / Methodology

MSstats Group Comparison (R)

library(MSstats)

comparison_matrix <- matrix(c(1, -1, 0, 0,
                               1, 0, -1, 0,
                               0, 1, -1, 0),
                             nrow = 3, byrow = TRUE)
rownames(comparison_matrix) <- c('Treatment1-Control', 'Treatment2-Control', 'Treatment1-Treatment2')
colnames(comparison_matrix) <- c('Control', 'Treatment1', 'Treatment2', 'Treatment3')

results <- groupComparison(contrast.matrix = comparison_matrix, data = processed)

sig_proteins <- results$ComparisonResult[results$ComparisonResult$adj.pvalue < 0.05 &
                                          abs(results$ComparisonResult$log2FC) > 1, ]

limma for Proteomics (R)

library(limma)

design <- model.matrix(~ 0 + condition, data = sample_info)
colnames(design) <- levels(sample_info$condition)

fit <- lmFit(protein_matrix, design)
contrast_matrix <- makeContrasts(Treatment - Control, levels = design)
fit2 <- contrasts.fit(fit, contrast_matrix)
fit2 <- eBayes(fit2)

results <- topTable(fit2, number = Inf, adjust.method = 'BH')
sig_results <- results[results$adj.P.Val < 0.05 & abs(results$logFC) > 1, ]

proDA (Missing Value-Aware)

library(QFeatures)
library(proDA)

fit <- proDA(protein_matrix, design = ~ condition, data = sample_info)
results <- test_diff(fit, contrast = 'conditionTreatment')
results$adj_pval <- p.adjust(results$pval, method = 'BH')
sig_results <- results[results$adj_pval < 0.05 & abs(results$diff) > 1, ]

Python: scipy/statsmodels

import pandas as pd
import numpy as np
from scipy import stats
from statsmodels.stats.multitest import multipletests

def differential_test(intensities, group1_cols, group2_cols):
    results = []
    for protein in intensities.index:
        g1 = intensities.loc[protein, group1_cols].dropna()
        g2 = intensities.loc[protein, group2_cols].dropna()

        if len(g1) >= 2 and len(g2) >= 2:
            stat, pval = stats.ttest_ind(g1, g2)
            log2fc = g2.mean() - g1.mean()
            results.append({'protein': protein, 'log2FC': log2fc, 'pvalue': pval})

    df = pd.DataFrame(results)
    df['adj_pvalue'] = multipletests(df['pvalue'], method='fdr_bh')[1]
    return df

sig = results[(results['adj_pvalue'] < 0.05) & (abs(results['log2FC']) > 1)]

Volcano Plot (R)

library(ggplot2)

ggplot(results, aes(x = log2FC, y = -log10(adj.P.Val))) +
    geom_point(aes(color = significant), alpha = 0.6) +
    geom_hline(yintercept = -log10(0.05), linetype = 'dashed') +
    geom_vline(xintercept = c(-1, 1), linetype = 'dashed') +
    scale_color_manual(values = c('grey', 'red')) +
    theme_minimal()

Parameters

--method

limma

--fdr-cutoff

0.05

--fc-cutoff

1.0

Why This Exists

• Without it: T-tests assume independence and fail on complex experimental designs or missing MS values
• With it: Validated mixed-models (MSstats) and empirical Bayes (Limma) correctly handle imputation and variances
• Why OmicsClaw: Standardizes notoriously strict R packages into a simple Python wrapper

Workflow

1. Calculate: Prepare contrast matrix and condition vectors.
2. Execute: Calculate log ratios and moderated t-statistics per protein.
3. Assess: Perform Benjamini-Hochberg FDR correction.
4. Generate: Output structured significant hits.
5. Report: Synthesize Volcano plots and heatmaps.

Example Queries

• "Run MSstats differential abundance on this data"
• "Find differentially expressed proteins using limma"

Output Structure

output_directory/
├── report.md
├── result.json
├── statistics.csv
├── figures/
│   └── volcano_plot.png
├── tables/
│   └── differential_results.csv
└── reproducibility/
    ├── commands.sh
    ├── environment.yml
    └── checksums.sha256

Safety

• Local-first: Strict offline processing without external upload.
• Disclaimer: Requires OmicsClaw reporting structures and disclaimers.
• Audit trail: Hyperparameters and operational flow states are logged fully.

Integration with Orchestrator

Trigger conditions:

• Automatically invoked dynamically based on tool metadata and user intent matching.

Chaining partners:

• quantification

  — Upstream normalized expression

• prot-enrichment

  — Downstream pathway enrichment

Version Compatibility

Reference examples tested with: limma 3.58+, MSstats 4.8+, scipy 1.12+, statsmodels 0.14+

Dependencies

Required: numpy, pandas, scipy, statsmodels Optional: MSstats (R), limma (R), proDA (R), QFeatures (R)

Citations

• MSstats — Choi et al., MCP 2014
• limma — Ritchie et al., Nucleic Acids Research 2015
• proDA — Ahlmann-Eltze & Anders, bioRxiv 2019

Related Skills

• quantification

  — Prepare normalized data for testing

• prot-enrichment

  — Pathway enrichment of significant proteins

• ptm

  — PTM-level differential analysis