Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|
A 4-week computational drug repurposing pipeline for Lung Adenocarcinoma (LUAD)
using TCGA RNA-seq data · DESeq2 · LINCS L1000 · CMap · KEGG/GO pathway analysis
Overview · Pipeline · Results · Files · How to Run · References
Lung Adenocarcinoma (LUAD) is the most common subtype of non-small cell lung cancer, with a 5-year survival rate below 20%. This project presents a complete AI-driven computational pipeline to identify existing approved drugs that can potentially be repurposed to treat LUAD — reducing the drug discovery timeline from 15 years → 4 weeks and cost from $2 billion → $50 million.
| Field | Details |
|---|---|
| Dataset | TCGA-LUAD (GDC Portal) |
| Samples | 583 total — 524 Tumor + 59 Normal |
| Genome Build | GRCh38/hg38 · Gencode v22 |
| Analysis Tools | DESeq2 (R), PyDESeq2 (Python), clusterProfiler, CMap clue.io |
| Duration | 1 month (20.04.2026 – 15.05.2026) |
| Organiser | BioCogniz — NGS · Drug Discovery · AI |
TCGA-LUAD RNA-seq (583 samples)
↓
QC + Normalization (DESeq2)
PCA: PC1=16%, PC2=9% — clear tumor/normal separation
↓
Differential Expression Analysis
5,756 DEGs | padj < 0.01 | |log₂FC| > 1.5
4,431 Upregulated · 1,325 Downregulated
↓
L1000 Landmark Gene Filtering
5,756 DEGs ∩ 978 L1000 genes = 102 signature genes
48 UP + 54 DOWN
↓
CMap Query (clue.io)
1,165,466 profiles searched | TAG score < −1.2
↓
Pathway Enrichment (KEGG + GO:BP)
Cell cycle #1 pathway (14/31 genes, p = 6.46×10⁻¹⁵)
Mitotic addiction confirmed (60% UP genes, p = 3.94×10⁻³¹)
↓
Literature Validation
PubMed · ClinicalTrials.gov · DrugBank
↓
10 Priority Drug Candidates
3 Clinically Validated · 2 FDA-Approved Recovered
| Metric | Value |
|---|---|
| Total genes tested | 39,044 |
| Significant DEGs | 5,756 |
| Upregulated | 4,431 (FAM83A +6.79, TOP2A +3.85, AURKA +3.20) |
| Downregulated | 1,325 (SEMA3G −3.23, EPAS1 −2.72, PECAM1 −2.32) |
| CMap signature | 102 genes (48 UP + 54 DOWN) |
| Hallmark | Evidence | p-value |
|---|---|---|
| Mitotic Addiction | 29/48 UP genes involved in nuclear division | 3.94×10⁻³¹ |
| Loss of Normal Identity | 18/54 DOWN genes in blood vessel development | 2.11×10⁻²¹ |
| Cell Cycle pathway | 14/31 annotated UP genes | 6.46×10⁻¹⁵ |
| Rank | Drug | TAG Score | Mechanism | Status |
|---|---|---|---|---|
| 1 | BAY-11-7085 | −1.71 | NF-kB inhibitor | Preclinical |
| 2 | KPT-185 | −1.59 | XPO1 antagonist | Phase I/II |
| 3 | bortezomib | −1.34 | Proteasome inhibitor | ⭐ FDA Approved |
| 6 | THZ-1 | −1.31 | CDK7 inhibitor | Nature 2014 ✓ |
| 7 | AZD-3463 | −1.31 | ALK inhibitor | ⭐ Phase I LUAD |
Pipeline Validation: AZD-3463 (ALK inhibitor) was independently recovered by the unbiased AI pipeline. ALK inhibitors are already FDA-approved first-line LUAD treatments — confirming the pipeline's biological validity.
LUAD-CMap-Drug-Repurposing/ │ ├── 📁 data/ │ ├── sample_metadata.csv # 583 sample info (tumor/normal labels) │ ├── DEG_results_all.csv # Full DESeq2 results (39,044 genes) │ ├── DEG_results_significant.csv # 5,756 significant DEGs │ ├── DEG_results_annotated.csv # DEGs with HGNC symbols │ ├── cmap_UP_genes.txt # 48 L1000 upregulated genes │ ├── cmap_DOWN_genes.txt # 54 L1000 downregulated genes │ ├── gene_signature_102.csv # Final CMap query signature │ └── cmap_results.csv # CMap connectivity scores │ ├── 📁 scripts/ │ ├── 01_data_download.py # TCGA-LUAD download via GDC API │ ├── 02_qc_normalization.py # Library size QC + DESeq2 normalization │ ├── 03_deseq2_analysis.py # Differential expression (PyDESeq2) │ ├── 04_l1000_filtering.py # Filter DEGs to L1000 landmark genes │ ├── 05_cmap_prep.py # Prepare CMap query input files │ ├── 06_pathway_analysis.R # KEGG + GO:BP enrichment (clusterProfiler) │ └── 07_visualization.R # Publication figures (ggplot2, EnhancedVolcano) │ ├── 📁 figures/ │ ├── PCA_LUAD.png # PCA plot — tumor vs normal separation │ ├── Volcano_LUAD.png # Volcano plot — 5,756 DEGs │ ├── Venn_L1000_filtering.png # Venn: 5,756 DEGs ∩ 978 L1000 = 102 genes │ ├── CMap_Drug_Ranking.png # Drug connectivity score bar chart │ ├── KEGG_UP_dotplot.png # KEGG enrichment — upregulated genes │ ├── KEGG_DOWN_dotplot.png # KEGG enrichment — downregulated genes │ └── DEG_Heatmap_Top50.png # Heatmap — top 50 DEGs │ ├── 📁 reports/ │ ├── LUAD_Research_Article.docx # Full 18-page research article │ └── LUAD_Presentation.pptx # 8-slide project presentation │ ├── requirements.txt # Python dependencies ├── R_packages.R # R package installation script └── README.md # This file
# Python 3.9+
pip install -r requirements.txt
# R 4.x — install packages
Rscript R_packages.R# Step 1: Download TCGA-LUAD data (requires internet, ~30–60 min)
python scripts/01_data_download.py
# Step 2: Quality control and normalization
python scripts/02_qc_normalization.py
# Step 3: Differential expression analysis
python scripts/03_deseq2_analysis.py
# Step 4: L1000 gene filtering
python scripts/04_l1000_filtering.py
# Step 5: Prepare CMap query files
python scripts/05_cmap_prep.py
# → Upload data/cmap_UP_genes.txt and data/cmap_DOWN_genes.txt
# to https://clue.io/query → download results → save as data/cmap_results.csv
# Step 6: Pathway enrichment analysis (R)
Rscript scripts/06_pathway_analysis.R
# Step 7: Generate all figures (R)
Rscript scripts/07_visualization.Rpydeseq2>=0.4.0 pandas>=1.5.0 numpy>=1.23.0 matplotlib>=3.6.0 seaborn>=0.12.0 scipy>=1.9.0 scikit-learn>=1.1.0 mygene>=3.2.2 requests>=2.28.0 openpyxl>=3.0.0 adjustText>=0.8
| Tool/Database | Purpose | Reference |
|---|---|---|
| TCGA-GDC | RNA-seq data source | NCI GDC portal |
| DESeq2 | Differential expression | Love et al., 2014 |
| CMap / clue.io | Drug connectivity query | Subramanian et al., 2017 |
| LINCS L1000 | Landmark gene filtering | LINCS program |
| clusterProfiler | KEGG / GO enrichment | Yu et al., 2012 |
| DrugBank | Drug target annotation | Wishart et al., 2018 |
| org.Hs.eg.db | Gene ID mapping | Bioconductor |
| PubMed | Literature validation | NCBI |
| Figure | Description |
|---|---|
| PCA_LUAD.png | Complete tumor-normal separation — PC1=16%, PC2=9% |
| Volcano_LUAD.png | 5,756 DEGs — FAM83A (top UP), SEMA3G (top DOWN) |
| Venn_L1000.png | 102-gene CMap signature construction |
| CMap_Drugs.png | 15 drug candidates ranked by TAG score |
| KEGG_dotplots.png | Cell cycle p=6.46×10⁻¹⁵ dominates UP enrichment |
- Love MI, et al. (2014). DESeq2. Genome Biology, 15, 550.
- Subramanian A, et al. (2017). CMap L1000. Cell, 171(6), 1437–1452.
- Lamb J, et al. (2006). Connectivity Map. Science, 313, 1929–1935.
- TCGA Research Network (2014). LUAD. Nature, 511, 543–550.
- Kwiatkowski N, et al. (2014). CDK7 inhibitor THZ-1. Nature, 511, 616–620.
- Yu G, et al. (2012). clusterProfiler. OMICS, 16(5), 284–287.
This project was completed as part of a 1-month AI in Drug Discovery internship at BioCogniz (20.04.2026 – 15.05.2026).
BioCogniz — NGS | Drug Discovery | AI
Made with ❤️ by Sangamithra J · Stella Maris College · 2026