An end-to-end bioinformatics pipeline for label-free quantitative (LFQ) proteomics analysis — from MaxQuant output to pathway-level biological interpretation.
This project was built to practice and demonstrate a real proteomics workflow using a publicly available neuronal dataset. The dataset comes from SH-SY5Y human neuroblastoma cells, a widely used model in neurodegeneration research, comparing three differentiation states: undifferentiated, RA-differentiated, and RA+PMA-differentiated.
I wanted to work through a complete proteomics analysis independently — not just run tools, but understand every step: why we impute missing values the way we do, why limma outperforms a simple t-test at low sample sizes, and how pathway enrichment connects protein-level changes to biology. The dataset was chosen specifically because the findings (cytoskeletal remodeling, synaptic pathways, axon guidance) are directly relevant to neuronal maintenance and neuromuscular research.
data/proteinGroups.txt
│
▼
scripts/proteomics_analysis.py
├── Filter contaminants and reverse hits
├── Log2 transformation of LFQ intensities
├── Perseus-style downshift imputation for missing values
├── PCA to check sample clustering
├── Welch's t-test + Benjamini-Hochberg FDR correction
└── Volcano plots and heatmaps
│
▼
scripts/limma_analysis.R
├── Empirical Bayes linear modeling (limma)
├── Two contrasts: RA vs Undiff, PMA vs Undiff
└── Volcano plots
│
▼
scripts/pathway_enrichment.py
├── Over-representation analysis via Enrichr (gseapy)
├── GO Biological Process, GO Cellular Component, KEGG, Reactome
├── Dotplots of top enriched terms
└── Filtered output highlighting neuronal and cytoskeletal pathways
proteomics_dia_analysis_pipeline/
├── data/
│ └── proteinGroups.txt # MaxQuant output — not tracked by git
├── env/
│ └── environment.yml # Conda environment
├── scripts/
│ ├── proteomics_analysis.py # Preprocessing, PCA, t-test, plots
│ ├── limma_analysis.R # Differential expression with limma
│ └── pathway_enrichment.py # GO/KEGG/Reactome enrichment
├── results/
│ ├── pca.png
│ ├── volcano_undiff_vs_RA.png
│ ├── volcano_undiff_vs_PMA.png
│ ├── volcano_limma_RA_vs_Undiff.png
│ ├── volcano_limma_PMA_vs_Undiff.png
│ ├── heatmap_undiff_vs_RA.png
│ ├── heatmap_undiff_vs_PMA.png
│ ├── MA_plot_limma.png
│ ├── significant_proteins_undiff_vs_RA.csv
│ ├── significant_proteins_undiff_vs_PMA.csv
│ ├── significant_proteins_limma_RA_vs_Undiff.csv
│ ├── significant_proteins_limma_PMA_vs_Undiff.csv
│ ├── all_proteins_undiff_vs_RA.csv
│ ├── all_proteins_undiff_vs_PMA.csv
│ ├── all_proteins_limma_RA_vs_Undiff.csv
│ ├── all_proteins_limma_PMA_vs_Undiff.csv
│ └── enrichment/
│ ├── dotplot_RA_vs_Undiff.png
│ ├── dotplot_PMA_vs_Undiff.png
│ ├── enrichment_RA_vs_Undiff.csv
│ ├── enrichment_PMA_vs_Undiff.csv
│ ├── DLR_relevant_pathways_RA_vs_Undiff.csv
│ └── DLR_relevant_pathways_PMA_vs_Undiff.csv
├── .gitignore
├── requirements.txt
├── LICENSE
└── README.md
git clone https://github.com/Kin-zala/proteomics_dia_analysis_pipeline.git
cd proteomics_dia_analysis_pipelineUsing conda (recommended):
conda env create -f env/environment.yml
conda activate proteomics_envOr using pip:
pip install -r requirements.txtif (!require("BiocManager")) install.packages("BiocManager")
BiocManager::install("limma")
install.packages(c("dplyr", "ggplot2"))Download proteinGroups.txt from PRIDE PXD031054 and place it in the data/ folder. The file is not tracked by git due to its size.
Run the three scripts in order:
Step 1 — Preprocessing and differential analysis
python scripts/proteomics_analysis.pyStep 2 — Limma differential expression
Rscript scripts/limma_analysis.RStep 3 — Pathway enrichment
python scripts/pathway_enrichment.pyAll outputs are saved automatically to the results/ folder.
| Comparison | Significant proteins (t-test) | Significant proteins (limma) |
|---|---|---|
| Undiff vs RA | 76 | 118 |
| Undiff vs PMA | 143 | 162 |
Limma consistently finds more significant hits than the t-test because it borrows variance information across all proteins (empirical Bayes smoothing), making it more sensitive — especially with n=6 per group.
Key pathways enriched (neuronal and cytoskeletal relevance):
| Pathway | Comparison | FDR |
|---|---|---|
| Cytoskeleton (GO:0005856) | Both | < 0.02 |
| Actin Cytoskeleton (GO:0015629) | PMA | < 0.02 |
| Regulation of actin cytoskeleton | Both | < 0.03 |
| Axon Guidance | Both | < 0.03 |
| GABAergic / Glutamatergic / Cholinergic synapse | RA | < 0.05 |
| Muscle Contraction | RA | < 0.04 |
| ATR Replication Stress Response | RA | < 0.01 |
The stronger cytoskeletal signal in the PMA condition compared to RA alone is consistent with RA+PMA being a more complete differentiation protocol — something visible both in the protein counts and the enrichment FDR values.
| Step | Method |
|---|---|
| Filtering | Remove reverse hits, contaminants, only-identified-by-site |
| Normalization | Log2 transformation of LFQ intensities |
| Missing value filter | Minimum 50% valid values per protein across all samples |
| Imputation | Perseus-style downshift (shift = 1.8 SD, width = 0.3 SD) |
| Differential expression | Welch's t-test + BH FDR (Python); limma eBayes (R) |
| Significance threshold | FDR < 0.05 and |log2FC| > 1 |
| Enrichment | Over-representation analysis via Enrichr (gseapy) |
| Databases | GO BP, GO CC, KEGG 2021, Reactome 2022 |
| Field | Detail |
|---|---|
| Source | PRIDE Archive |
| Accession | PXD031054 |
| Organism | Homo sapiens |
| Cell line | SH-SY5Y neuroblastoma |
| Conditions | Undifferentiated, RA, RA+PMA |
| Replicates | n = 6 per condition |
| Format | MaxQuant proteinGroups.txt (LFQ) |
Kinnari Zala
MSc Bioinformatics
github.com/Kin-zala
This project is licensed under the MIT License — see LICENSE for details.