| name | proteomics |
| description | Proteomics analysis covering DDA, DIA, TMT/iTRAQ, Olink, and SomaScan platforms including preprocessing, normalization, differential expression, and PTM analysis |
| version | 1.0.0 |
| tags | ["proteomics","dda","dia","tmt","olink","somascan","mass-spectrometry"] |
Proteomics Analysis
Method selection and execution guidance for quantitative proteomics across mass spectrometry and affinity-based platforms.
Platform Detection and Preprocessing
Identify the platform first — each has distinct data characteristics and preprocessing requirements:
Platform?
├── DDA (Data-Dependent Acquisition)
│ ├── Input: FragPipe/MaxQuant output → protein/peptide intensity matrix
│ ├── Preprocessing:
│ │ ├── 1. Filter by missingness (keep proteins detected in >= 70% of at least one group)
│ │ ├── 2. Log2 transform intensities
│ │ ├── 3. Normalize: VSN (variance stabilizing, preferred) or median centering
│ │ ├── 4. Impute missing values:
│ │ │ ├── MNAR (Missing Not At Random, below LOD) → MinProb (left-censored)
│ │ │ ├── MAR (Missing At Random, technical dropout) → kNN
│ │ │ └── Mixed → hybrid (MinProb for low-abundance, kNN for rest)
│ │ └── 5. Assess batch effects (PCA by batch)
│ └── DE: DEP test_diff (default) or limma via rpy2
│
├── DIA (Data-Independent Acquisition)
│ ├── Input: DIA-NN output → protein/precursor intensity matrix
│ ├── Preprocessing: Same as DDA but fewer missing values expected
│ │ ├── DIA-NN already provides normalized intensities (check MaxLFQ column)
│ │ └── If raw intensities: filter → log2 → normalize → impute as above
│ └── DE: DEP test_diff or limma via rpy2
│
├── TMT / iTRAQ (Isobaric Labeling)
│ ├── Input: FragPipe/MaxQuant PSM-level or protein-level quantification
│ ├── Preprocessing:
│ │ ├── 1. MSstatsTMT dataProcess (handles plex normalization, protein summarization)
│ │ ├── 2. Channel-level normalization (reference channel, global median)
│ │ └── 3. Between-plex normalization (bridge channel if available)
│ └── DE: MSstatsTMT groupComparison (handles plex as random effect)
│
├── Olink (Proximity Extension Assay)
│ ├── Input: NPX values (already log2-scale, plate-normalized)
│ ├── Preprocessing:
│ │ ├── 1. LOD filtering: flag assays below LOD per sample
│ │ ├── 2. Multi-plate: bridge normalization using bridge samples
│ │ ├── 3. QC warnings: remove samples/assays with QC_Warning flags
│ │ └── 4. No additional normalization needed (NPX is pre-normalized)
│ └── DE: limma via rpy2 (treats NPX as log2-intensity) or linear model
│
└── SomaScan (SOMAmer Aptamer)
├── Input: ADAT file → RFU (Relative Fluorescence Units) matrix
├── Preprocessing:
│ ├── 1. ANML normalization (SomaScan standard, usually pre-applied)
│ ├── 2. Log2 transform RFU values
│ ├── 3. Filter: remove aptamers with high CV across controls
│ └── 4. Plate/batch correction if multi-plate
└── DE: limma via rpy2 on log2 RFU values
Differential Expression Method Selection
Design complexity?
├── Simple (2 groups, no covariates)
│ ├── DDA/DIA → DEP test_diff (R via rpy2, wrapper around limma)
│ └── TMT → MSstatsTMT groupComparison
├── Complex (multiple factors, covariates, interactions)
│ ├── DDA/DIA/Olink/SomaScan → limma via rpy2 (full formula support)
│ └── TMT → MSstats groupComparison with contrast matrix
├── Longitudinal / repeated measures
│ └── dream (variancePartition) via rpy2 (mixed-effects model)
└── Paired samples
└── limma with blocking factor or paired design matrix
PTM Analysis (Phosphoproteomics)
Phosphoproteomics workflow:
├── Quantification: Site-level intensities (NOT protein rollup)
│ ├── FragPipe: STY_79.9663 modification
│ └── MaxQuant: Phospho (STY)Sites table
├── Normalization: Normalize to total proteome if available (correct for protein abundance changes)
├── DE: limma on log2 site-level intensities
├── Kinase activity: decoupler with a kinase-substrate network you resolved
│ ├── dc.mt.ulm(adata, net=ksn) where ksn = kinase-substrate network
│ └── NOTE: no kinase-substrate network is in the reference inventory, and
│ every decoupler built-in fetcher (dc.op.*) reaches the network, which
│ is blocked — so Omnipath/PhosphoSitePlus retrieval fails outright.
│ If none is available, report it and deliver site-level DE without
│ kinase activity; never invent a path or substitute a TF regulon.
└── Critical: Do NOT roll up phospho-sites to protein level — this loses site-specific information
Preprocessing Workflow
- Load data: Read platform-specific output format into pandas DataFrame.
- Filter: Remove contaminants, reverse hits (MaxQuant), low-detection proteins.
- Transform: Log2 intensities (skip for Olink NPX, already log2).
- Normalize: VSN (DDA/DIA), median (DDA/DIA alternative), MSstatsTMT (TMT), bridge (Olink multi-plate). Do NOT normalize SomaScan if ANML already applied.
- Impute: Assess missing value pattern (MNAR vs MAR). Apply appropriate imputation.
- Batch check: PCA colored by batch/plate/plex. Correct if needed (ComBat or include as covariate).
- DE testing: Apply method from decision tree. Extract results with log2FC, p-value, adjusted p-value.
- Downstream: Pathway enrichment via decoupler, protein-protein interaction networks.
Output Conventions
- Save processed protein matrix as CSV: rows = proteins (UniProt ID + gene symbol), columns = samples.
- DE results as CSV: protein, gene, log2FoldChange, pvalue, padj, avg_intensity.
- Volcano plot (log2FC vs -log10 padj) and intensity distribution boxplots as PNG/PDF.
- Missing value heatmap showing pattern (MNAR vs MAR assessment).
- PCA plot of samples colored by condition and batch/plate.
- Protein ID mapping table: UniProt accession to gene symbol.
Anti-Patterns
- Treating protein intensities like RNA-seq counts: Protein intensities are continuous, not discrete counts. Do NOT use DESeq2/edgeR/PyDESeq2 on proteomics data. Use limma, DEP, or MSstats.
- Ignoring missing value patterns: MNAR (below detection limit) and MAR (random dropout) require different imputation strategies. Using kNN for MNAR or MinProb for MAR introduces systematic bias.
- Not checking for batch effects in multi-plate studies: Olink multi-plate, TMT multi-plex, and DDA multi-batch experiments all require explicit batch correction. Uncorrected batch effects dominate real biology.
- Protein-level analysis on phosphoproteomics data: Rolling up phospho-sites to protein level destroys site-specific regulation. A protein can have sites with opposite regulation (activating vs inhibiting phosphorylation).
- Normalizing already-normalized data: Olink NPX values are pre-normalized. SomaScan with ANML is pre-normalized. Double-normalization distorts the data.
- Imputing before filtering: Impute after removing low-quality proteins/samples. Imputing values for proteins with >50% missing is unreliable.
- Using t-tests without multiple testing correction: Always apply BH/FDR correction. With thousands of proteins, uncorrected p-values produce massive false positives.
- Ignoring protein grouping: MaxQuant groups proteins sharing peptides. Using individual protein IDs without resolving groups inflates the number of "hits" and introduces redundancy.
Additional Available Packages
Python
- spectrum-utils: Mass spectrometry spectrum visualization — annotated spectra, mirror plots for spectral library matching. Use for QC and figure generation.
- ms-deisotope: Deisotoping and charge state deconvolution. Use for preprocessing high-resolution MS1/MS2 spectra before database search.
R (via rpy2)
- MSnbase: Legacy MS data class infrastructure.
readMSData() for reading mzML. Foundation for xcms.
- Spectra (Bioconductor): Modern MS data backend replacing MSnbase. Use
Spectra(), filterMsLevel(), peaksData().
References
API references for all supported packages:
references/dep-rpy2-api.md — DEP for DDA/DIA proteomics (filtering, normalization, imputation, DE)references/msstats-rpy2-api.md — MSstats/MSstatsTMT for label-free and TMT proteomicsreferences/olink-somascan-api.md — Olink NPX and SomaScan RFU preprocessing patternsreferences/pyteomics-api.md — Python toolkit for reading MS formats (mzML, pepXML)
