Proteomics Analysis

Comprehensive analysis of mass spectrometry-based proteomics data from protein identification through quantification, differential expression, post-translational modifications, and systems-level interpretation.

When to Use This Skill

Triggers:

• User has proteomics data (MS output files)
• Questions about protein abundance or expression
• Differential protein expression analysis requests
• PTM analysis (phosphorylation, acetylation, ubiquitination)
• Protein-RNA correlation analysis
• Multi-omics integration involving proteomics
• Protein complex or interaction analysis
• Proteomics biomarker discovery

Example Questions This Skill Solves:

1. "Analyze this MaxQuant output for differential protein expression"
2. "Which proteins are significantly upregulated in disease vs control?"
3. "Correlate protein abundance with mRNA expression"
4. "What post-translational modifications change between conditions?"
5. "Identify protein complexes in my co-IP MS data"
6. "Which pathways are enriched in differentially expressed proteins?"
7. "Find protein biomarkers for disease classification"
8. "Compare protein and RNA levels to identify translation-regulated genes"

Core Capabilities

| Capability | Description | |-----------|-------------| | Data Import | MaxQuant, Spectronaut, DIA-NN, Proteome Discoverer, FragPipe outputs | | Quality Control | Missing value analysis, intensity distributions, sample clustering | | Normalization | Median, quantile, TMM, VSN normalization methods | | Imputation | MinProb, KNN, QRILC for missing values | | Differential Expression | Limma, DEP, MSstats for statistical testing | | PTM Analysis | Phospho-site localization, PTM enrichment, kinase prediction | | Protein-RNA Integration | Correlation analysis, translation efficiency | | Pathway Enrichment | Over-representation and GSEA for protein sets | | PPI Analysis | Protein complex detection, interaction networks via STRING/IntAct | | Reporting | Comprehensive reports with volcano plots, heatmaps, pathway diagrams |

Workflow Overview

Input: MS Proteomics Data
    |
    v
Phase 1: Data Import & QC
    |-- Load MaxQuant/Spectronaut/DIA-NN output
    |-- Parse protein groups, intensities, modifications
    |-- Quality control plots (missing values, intensity distributions)
    |-- Sample correlation and PCA
    |
    v
Phase 2: Preprocessing
    |-- Filter low-confidence proteins
    |-- Handle missing values (imputation or filtering)
    |-- Log-transform intensities
    |-- Normalize across samples
    |
    v
Phase 3: Differential Expression Analysis
    |-- Statistical testing (limma, t-test, ANOVA)
    |-- Multiple testing correction (BH, Bonferroni)
    |-- Fold change calculation
    |-- Significance thresholds (p < 0.05, |log2FC| > 1)
    |
    v
Phase 4: PTM Analysis (if applicable)
    |-- Identify modified peptides
    |-- Localization probability filtering
    |-- PTM site quantification
    |-- Kinase-substrate prediction
    |-- PTM enrichment analysis
    |
    v
Phase 5: Functional Enrichment
    |-- Gene Ontology enrichment
    |-- KEGG/Reactome pathway enrichment
    |-- Protein complex enrichment (CORUM)
    |-- Tissue-specific enrichment
    |
    v
Phase 6: Protein-Protein Interactions
    |-- Query STRING for interaction networks
    |-- Identify protein complexes
    |-- Network clustering and modules
    |-- Hub protein identification
    |
    v
Phase 7: Multi-Omics Integration (optional)
    |-- Correlate with RNA-seq data
    |-- Identify translation-regulated proteins
    |-- Compare with variant/CNV data
    |-- Integrate with metabolomics
    |
    v
Phase 8: Generate Report
    |-- Summary statistics
    |-- Volcano plots and heatmaps
    |-- Pathway diagrams
    |-- Protein network visualizations
    |-- Multi-omics integration plots

Phase Details

Phase 1: Data Import & Quality Control

Objective: Load proteomics data and assess data quality.

Supported input formats:

MaxQuant (most common):

• proteinGroups.txt - Protein-level quantification
• evidence.txt - Peptide-level data
• Phospho (STY)Sites.txt - Phosphorylation sites
• modificationSpecificPeptides.txt - Other PTMs

Spectronaut:

• *_Report.tsv - Protein/peptide quantification
• DIA-based quantification

DIA-NN:

• report.tsv - Protein groups
• report.pr_matrix.tsv - Protein matrix

Proteome Discoverer:

• *_Proteins.txt
• *_PSMs.txt

Data loading:

python
def load_maxquant_proteins(protein_groups_file):
    """
    Load MaxQuant proteinGroups.txt file.

    Returns:
    - DataFrame with proteins as rows, samples as columns
    - Metadata (protein names, gene names, sequence coverage)
    """
    import pandas as pd

    # Read file
    df = pd.read_csv(protein_groups_file, sep='\t')

    # Extract intensity columns (LFQ or raw)
    intensity_cols = [col for col in df.columns if 'LFQ intensity' in col or 'Intensity ' in col]

    # Create intensity matrix
    intensity_matrix = df[intensity_cols].copy()
    intensity_matrix.columns = [col.replace('LFQ intensity ', '').replace('Intensity ', '')
                                 for col in intensity_cols]

    # Add protein metadata
    metadata = df[['Protein IDs', 'Gene names', 'Fasta headers',
                   'Peptides', 'Sequence coverage [%]']].copy()

    return intensity_matrix, metadata

Quality Control:

1. Missing value assessment:

python
def assess_missing_values(intensity_matrix):
    """
    Calculate percentage of missing values per protein and sample.
    """
    # Per protein
    missing_per_protein = (intensity_matrix == 0).sum(axis=1) / intensity_matrix.shape[1]

    # Per sample
    missing_per_sample = (intensity_matrix == 0).sum(axis=0) / intensity_matrix.shape[0]

    # Visualize
    plot_missing_value_heatmap(intensity_matrix)

    return missing_per_protein, missing_per_sample

1. Intensity distribution:

python
def plot_intensity_distributions(intensity_matrix):
    """
    Plot log10 intensity distributions per sample.
    Check for consistent distributions across samples.
    """
    import matplotlib.pyplot as plt
    import numpy as np

    log_intensities = np.log10(intensity_matrix.replace(0, np.nan))

    # Boxplot per sample
    log_intensities.plot(kind='box')
    plt.ylabel('log10 Intensity')
    plt.title('Intensity Distribution per Sample')

    # Should see similar median and spread across samples

1. Sample correlation:

python
def plot_sample_correlation(intensity_matrix):
    """
    Calculate and visualize sample-sample correlation.
    Expect: High correlation within replicates, lower between conditions.
    """
    # Log-transform and remove zeros
    log_data = np.log2(intensity_matrix.replace(0, np.nan))

    # Correlation matrix
    corr_matrix = log_data.corr(method='pearson')

    # Heatmap
    import seaborn as sns
    sns.heatmap(corr_matrix, annot=True, cmap='RdYlBu_r', vmin=0.8, vmax=1.0)

1. PCA:

python
def perform_pca(intensity_matrix, sample_groups):
    """
    Principal component analysis for sample clustering.
    """
    from sklearn.decomposition import PCA

    # Prepare data (log, impute, scale)
    log_data = np.log2(intensity_matrix.replace(0, np.nan))
    # Simple imputation with minimum value
    imputed = log_data.fillna(log_data.min().min())

    # PCA
    pca = PCA(n_components=2)
    pca_result = pca.fit_transform(imputed.T)

    # Plot with group colors
    plt.scatter(pca_result[:, 0], pca_result[:, 1], c=sample_groups)
    plt.xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%})')
    plt.ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%})')

Phase 2: Preprocessing & Normalization

Objective: Clean data and normalize across samples for fair comparison.

Filtering:

python
def filter_proteins(intensity_matrix, metadata, min_valid=3):
    """
    Filter out low-confidence proteins.

    Criteria:
    - At least 2 unique peptides (from metadata)
    - At least min_valid samples with detected intensity
    - Remove contaminants and reverse sequences
    """
    # Filter by peptide count
    valid_proteins = metadata['Peptides'] >= 2

    # Filter by detection in samples
    n_detected = (intensity_matrix > 0).sum(axis=1)
    valid_detection = n_detected >= min_valid

    # Remove contaminants (from MaxQuant)
    is_contaminant = metadata['Protein IDs'].str.contains('CON__', na=False)
    is_reverse = metadata['Protein IDs'].str.contains('REV__', na=False)

    # Combined filter
    keep = valid_proteins & valid_detection & ~is_contaminant & ~is_reverse

    return intensity_matrix[keep], metadata[keep]

Missing value imputation:

python
def impute_missing_values(intensity_matrix, method='MinProb'):
    """
    Impute missing protein intensities.

    Methods:
    - MinProb: Random from minimum observed + normal noise (for MNAR)
    - KNN: K-nearest neighbors imputation
    - QRILC: Quantile regression-based imputation
    """
    if method == 'MinProb':
        # Assume missing = low abundance (MNAR assumption)
        min_val = intensity_matrix[intensity_matrix > 0].min().min()
        width = 0.3  # Standard deviation of noise
        shift = 1.8  # Downshift from minimum

        # Replace zeros with random low values
        imputed = intensity_matrix.copy()
        missing_mask = imputed == 0
        n_missing = missing_mask.sum().sum()

        random_vals = np.random.normal(
            loc=min_val - shift,
            scale=width,
            size=n_missing
        )
        imputed.values[missing_mask.values] = random_vals

        return imputed

    elif method == 'KNN':
        from sklearn.impute import KNNImputer
        imputer = KNNImputer(n_neighbors=5)
        imputed = pd.DataFrame(
            imputer.fit_transform(intensity_matrix.replace(0, np.nan)),
            index=intensity_matrix.index,
            columns=intensity_matrix.columns
        )
        return imputed

Normalization:

python
def normalize_intensities(intensity_matrix, method='median'):
    """
    Normalize protein intensities across samples.

    Methods:
    - median: Divide by median intensity per sample
    - quantile: Quantile normalization (same distribution)
    - TMM: Trimmed mean of M-values (from edgeR)
    - VSN: Variance-stabilizing normalization
    """
    if method == 'median':
        # Median normalization
        medians = intensity_matrix.median(axis=0)
        global_median = medians.median()
        norm_factors = global_median / medians
        normalized = intensity_matrix * norm_factors
        return normalized

    elif method == 'quantile':
        # Quantile normalization
        from sklearn.preprocessing import quantile_transform
        normalized = pd.DataFrame(
            quantile_transform(intensity_matrix, axis=1),
            index=intensity_matrix.index,
            columns=intensity_matrix.columns
        )
        return normalized

Phase 3: Differential Expression Analysis

Objective: Identify proteins with significant abundance changes between conditions.

Statistical testing with limma:

python
def differential_expression_limma(log2_intensities, group1_samples, group2_samples):
    """
    Perform differential expression using limma-like approach.

    Returns:
    - log2 fold changes
    - p-values
    - adjusted p-values (BH)
    """
    from scipy import stats

    results = []

    for protein in log2_intensities.index:
        # Extract intensities for each group
        group1 = log2_intensities.loc[protein, group1_samples]
        group2 = log2_intensities.loc[protein, group2_samples]

        # Calculate statistics
        mean1 = group1.mean()
        mean2 = group2.mean()
        log2fc = mean2 - mean1

        # t-test
        t_stat, p_value = stats.ttest_ind(group1, group2, equal_var=False)

        results.append({
            'protein': protein,
            'log2FC': log2fc,
            'mean_group1': mean1,
            'mean_group2': mean2,
            'p_value': p_value,
            't_statistic': t_stat
        })

    results_df = pd.DataFrame(results)

    # Multiple testing correction (Benjamini-Hochberg)
    from statsmodels.stats.multitest import multipletests
    results_df['adj_p_value'] = multipletests(results_df['p_value'], method='fdr_bh')[1]

    # Classify significance
    results_df['significant'] = (
        (results_df['adj_p_value'] < 0.05) &
        (np.abs(results_df['log2FC']) > 1.0)
    )

    return results_df

Volcano plot:

python
def plot_volcano(de_results, title='Volcano Plot'):
    """
    Visualize differential expression results.
    """
    import matplotlib.pyplot as plt

    plt.figure(figsize=(8, 6))

    # Non-significant
    non_sig = de_results[~de_results['significant']]
    plt.scatter(non_sig['log2FC'], -np.log10(non_sig['p_value']),
                c='gray', alpha=0.5, s=10)

    # Significant
    sig = de_results[de_results['significant']]
    plt.scatter(sig['log2FC'], -np.log10(sig['p_value']),
                c='red', alpha=0.7, s=20)

    # Thresholds
    plt.axhline(-np.log10(0.05), color='blue', linestyle='--', label='p=0.05')
    plt.axvline(-1, color='blue', linestyle='--')
    plt.axvline(1, color='blue', linestyle='--', label='|log2FC|=1')

    plt.xlabel('log2 Fold Change')
    plt.ylabel('-log10(p-value)')
    plt.title(title)
    plt.legend()

Phase 4: PTM Analysis

Objective: Analyze post-translational modifications (phosphorylation, acetylation, etc.)

Phosphoproteomics workflow:

python
def analyze_phosphosites(phospho_sites_file, intensity_matrix):
    """
    Analyze phosphorylation site changes.

    Input: MaxQuant Phospho (STY)Sites.txt
    Output: Differential phosphorylation per site
    """
    # Load phospho data
    phospho = pd.read_csv(phospho_sites_file, sep='\t')

    # Filter by localization probability
    phospho_confident = phospho[phospho['Localization prob'] > 0.75]

    # Extract site information
    phospho_confident['site'] = (
        phospho_confident['Gene names'] + '_' +
        phospho_confident['Amino acid'] +
        phospho_confident['Position'].astype(str)
    )

    # Quantification (similar to protein-level analysis)
    # ... perform differential analysis ...

    return phospho_results

Kinase-substrate prediction:

python
def predict_kinases(phospho_sites):
    """
    Predict upstream kinases for phosphorylation sites.

    Uses ToolUniverse PhosphoSitePlus or KEA3 tools.
    """
    from tooluniverse import ToolUniverse
    tu = ToolUniverse()

    # For each significant phosphosite
    kinase_predictions = []
    for site in phospho_sites:
        # Query kinase-substrate databases
        # (would use actual ToolUniverse tool here)
        result = tu.run_one_function({
            "name": "phosphosite_plus_query",  # hypothetical
            "arguments": {"site": site}
        })
        kinase_predictions.append(result)

    return kinase_predictions

Phase 5: Functional Enrichment

Objective: Interpret biological meaning of protein changes via pathway analysis.

Gene Ontology enrichment:

python
def pathway_enrichment_proteins(de_proteins, organism='human'):
    """
    Perform pathway enrichment for differentially expressed proteins.

    Uses ToolUniverse gene-enrichment skill.
    """
    from tooluniverse import ToolUniverse
    tu = ToolUniverse()

    # Extract gene names for significant proteins
    sig_proteins = de_proteins[de_proteins['significant']]
    gene_list = sig_proteins['gene_name'].tolist()

    # Run enrichment via ToolUniverse
    enrichment = tu.run_one_function({
        "name": "enrichr_enrich",
        "arguments": {
            "gene_list": ",".join(gene_list),
            "library": "KEGG_2021_Human"
        }
    })

    return enrichment

Protein complex enrichment:

python
def protein_complex_enrichment(protein_list):
    """
    Test for enrichment of known protein complexes (CORUM database).
    """
    # Query CORUM or use ToolUniverse
    # Identify if proteins are part of known complexes
    pass

Phase 6: Protein-Protein Interactions

Objective: Identify interaction networks and protein complexes.

STRING network analysis:

python
def build_protein_network(protein_list, confidence=0.7):
    """
    Build PPI network using STRING database.

    Uses ToolUniverse STRING tools.
    """
    from tooluniverse import ToolUniverse
    tu = ToolUniverse()

    # Get interactions
    interactions = tu.run_one_function({
        "name": "string_get_interactions",
        "arguments": {
            "proteins": ",".join(protein_list),
            "species": 9606,  # human
            "score_threshold": int(confidence * 1000)
        }
    })

    # Build network graph
    import networkx as nx
    G = nx.Graph()

    for interaction in interactions['data']:
        G.add_edge(
            interaction['protein1'],
            interaction['protein2'],
            score=interaction['score']
        )

    return G

Module detection:

python
def detect_protein_modules(network_graph):
    """
    Identify tightly connected protein modules (complexes).
    """
    from networkx.algorithms import community

    # Detect communities
    communities = community.greedy_modularity_communities(network_graph)

    # Annotate modules with enriched functions
    modules = []
    for i, comm in enumerate(communities):
        module_proteins = list(comm)
        # Run enrichment for this module
        enrichment = pathway_enrichment_proteins(module_proteins)
        modules.append({
            'module_id': i,
            'proteins': module_proteins,
            'size': len(module_proteins),
            'top_function': enrichment['top_terms'][0]
        })

    return modules

Phase 7: Multi-Omics Integration

Objective: Integrate proteomics with transcriptomics and other omics.

Protein-RNA correlation:

python
def correlate_protein_rna(protein_data, rna_data, common_samples):
    """
    Correlate protein and mRNA levels for each gene.

    Expected: r ~ 0.4-0.6 (moderate correlation)
    Discordance indicates post-transcriptional regulation
    """
    from scipy.stats import spearmanr

    # Find common genes
    common_genes = set(protein_data.index) & set(rna_data.index)

    correlations = {}
    for gene in common_genes:
        protein = protein_data.loc[gene, common_samples]
        rna = rna_data.loc[gene, common_samples]

        r, p = spearmanr(protein, rna)
        correlations[gene] = {
            'r': r,
            'p': p,
            'regulation': classify_regulation(r, protein.mean(), rna.mean())
        }

    return correlations

def classify_regulation(r, protein_level, rna_level):
    """
    Classify regulatory mechanism based on correlation and levels.
    """
    if r > 0.6 and protein_level > 0 and rna_level > 0:
        return 'transcriptional_upregulation'
    elif r > 0.6 and protein_level < 0 and rna_level < 0:
        return 'transcriptional_downregulation'
    elif r < 0.2 and protein_level > 0 and rna_level < 0:
        return 'translational_upregulation'
    elif r < 0.2 and protein_level < 0 and rna_level > 0:
        return 'protein_degradation'
    else:
        return 'mixed_regulation'

Integration with multi-omics skill:

python
def integrate_with_multiomics(protein_data, rna_data, methylation_data):
    """
    Pass proteomics data to multi-omics integration skill.

    Enables comprehensive analysis across all molecular layers.
    """
    # Prepare for multi-omics skill
    omics_data = {
        'proteomics': protein_data,
        'rnaseq': rna_data,
        'methylation': methylation_data
    }

    # Invoke multi-omics integration skill
    from tooluniverse import ToolUniverse
    # (Would use Skill tool to invoke tooluniverse-multi-omics-integration)

    return integrated_analysis

Phase 8: Report Generation

Generate comprehensive proteomics report:

markdown
# Proteomics Analysis Report

## Dataset Summary
- **Samples**: 20 (10 disease, 10 control)
- **Proteins Identified**: 5,432
- **Proteins Quantified**: 4,987 (at least 3 samples)
- **Platform**: Orbitrap Fusion Lumos, MaxQuant 2.0

## Quality Control
- **Missing Values**: 15% average per protein
- **Sample Correlation**: 0.92-0.98 within groups
- **PCA**: Clear separation between disease and control (PC1: 35% variance)

## Differential Expression
- **Significant Proteins**: 432 (adj. p < 0.05, |log2FC| > 1)
  - Upregulated: 245 proteins
  - Downregulated: 187 proteins
- **Top upregulated**: MYC (log2FC=3.2), EGFR (log2FC=2.8)
- **Top downregulated**: TP53 (log2FC=-2.5), BRCA1 (log2FC=-2.1)

## Phosphoproteomics
- **Phosphosites Quantified**: 8,543
- **Differentially Phosphorylated**: 234 sites (p < 0.05)
- **Top Predicted Kinases**: CDK1, MAPK1, AKT1

## Pathway Enrichment
### Top Pathways (Upregulated)
1. **Cell Cycle** (p=1e-15) - 45 proteins, including cyclins, CDKs
2. **DNA Replication** (p=1e-12) - 23 proteins
3. **Glycolysis** (p=1e-10) - 18 proteins

### Top Pathways (Downregulated)
1. **Apoptosis** (p=1e-14) - 32 proteins, including caspases
2. **DNA Repair** (p=1e-11) - 28 proteins
3. **Oxidative Phosphorylation** (p=1e-9) - 25 proteins

## Protein Network Analysis
- **Network**: 432 nodes, 1,245 edges (STRING confidence > 0.7)
- **Modules Detected**: 8 functional modules
  - Module 1: Cell cycle (85 proteins)
  - Module 2: Metabolism (62 proteins)
  - Module 3: Translation (48 proteins)

## Protein-RNA Correlation
- **Overall Correlation**: r = 0.54 (moderate, expected)
- **High Correlation**: 2,134 genes (r > 0.6) - transcriptional regulation
- **Low Correlation**: 456 genes (r < 0.2) - post-transcriptional regulation
- **Translation-Regulated**: 89 proteins (high protein, low RNA)

## Biological Interpretation
Disease state shows increased proliferation (MYC, cyclins) with concurrent
suppression of apoptosis and DNA repair (TP53, BRCA1). Metabolic shift toward
glycolysis evident at protein level. Post-transcriptional upregulation of
translation machinery suggests adaptation to proliferative demands.

## Potential Biomarkers
Top 10 proteins for disease classification (Random Forest AUC=0.95):
1. MYC (protein)
2. EGFR (protein)
3. CDK1 (phospho-T161)
4. TP53 (protein)
5. BRCA1 (protein)

Integration with ToolUniverse

Skills Coordinated: | Skill | Used For | Phase | |-------|----------|-------| | tooluniverse-gene-enrichment | Pathway enrichment | Phase 5 | | tooluniverse-protein-interactions | PPI networks | Phase 6 | | tooluniverse-rnaseq-deseq2 | RNA-seq for integration | Phase 7 | | tooluniverse-multi-omics-integration | Cross-omics analysis | Phase 7 | | tooluniverse-target-research | Protein annotation | Phase 8 |

Example Use Cases

Use Case 1: Cancer Proteomics

Question: "Analyze proteomics data from breast cancer vs normal tissue"

Workflow:

1. Load MaxQuant proteinGroups.txt
2. QC and filter (keep proteins with 2+ peptides, detected in 3+ samples)
3. Impute missing, normalize by median
4. Differential expression (limma): 432 significant proteins
5. Pathway enrichment: Cell cycle, metabolism upregulated
6. STRING network: Identify hub proteins (MYC, EGFR)
7. Integrate with TCGA RNA-seq: Find translation-regulated genes
8. Report: Comprehensive analysis with biomarkers

Use Case 2: Phosphoproteomics Signaling

Question: "What kinase signaling is activated in response to drug treatment?"

Workflow:

1. Load Phospho (STY)Sites.txt from MaxQuant
2. Filter by localization probability > 0.75
3. Differential phosphorylation analysis
4. Kinase prediction for significant sites
5. Identify MAPK1, CDK1, AKT1 as top kinases
6. Pathway enrichment: MAPK, PI3K/AKT pathways
7. Report: Drug activates growth signaling

Use Case 3: Protein-RNA Integration

Question: "Which proteins are regulated post-transcriptionally?"

Workflow:

1. Load proteomics (MaxQuant) and RNA-seq (DESeq2) data
2. Match samples, extract common genes
3. Correlate protein and RNA for each gene
4. Identify low-correlation genes (r < 0.2)
5. Classify: translation upregulation, protein degradation
6. Enrichment: Find pathways enriched in post-transcriptional regulation
7. Report: 89 translation-regulated proteins, RNA-binding proteins enriched

Quantified Minimums

| Component | Requirement | |-----------|-------------| | Proteins quantified | At least 500 proteins | | Replicates | At least 3 per condition | | Filtering | 2+ unique peptides per protein | | Statistical test | limma or t-test with multiple testing correction | | Pathway enrichment | At least one method (GO, KEGG, or Reactome) | | Report | Summary, QC, DE results, pathways, visualizations |

Limitations

• Platform-specific: Optimized for MS-based proteomics (not Western blot quantification)
• Missing values: High missing rate (>50% per protein) limits statistical power
• PTM analysis: Requires enrichment protocols for comprehensive PTM profiling
• Absolute quantification: Relative abundance only (unless TMT/SILAC used)
• Protein isoforms: Typically collapsed to gene level
• Dynamic range: MS has limited dynamic range vs mRNA sequencing

References

Methods:

• MaxQuant: https://doi.org/10.1038/nbt.1511
• Limma for proteomics: https://doi.org/10.1093/nar/gkv007
• DEP workflow: https://doi.org/10.1038/nprot.2018.107

Databases:

• STRING: https://string-db.org
• PhosphoSitePlus: https://www.phosphosite.org
• CORUM: https://mips.helmholtz-muenchen.de/corum