Drug Repurposing Through Disease Similarity Analysis Using BiomarkerKB

Computational pipeline and interactive web dashboard for identifying drug repurposing candidates across multiple cancers, driven by biomarker evidence from BiomarkerKB and cross-referenced against GTEx, Pharos/IDG, GDC/TCGA, LINCS L1000, and STRING.

Presentation: Google Slides

Full project description on Google Docs

---

Overview

The pipeline queries six public databases to build a ranked list of drug repurposing candidates for a given cancer:

Step	Source	What it provides
1	BiomarkerKB	Disease-associated biomarker genes
2	GTEx	Tissue expression (TPM) per gene
3	Pharos / IDG	Target development level (Tclin → Tdark)
4	GDC / TCGA	Somatic mutation frequency + differential expression vs. normal
5	STRING	Protein–protein interaction network expansion
6	LINCS L1000	Drug perturbagen signatures matching query genes

Candidates are scored across five weighted components and ranked. Weights can be adjusted interactively in the web dashboard.

Diseases supported (configurable): Hepatocellular Carcinoma, Pancreatic Cancer, Colorectal Cancer, Lung Adenocarcinoma and Breast Cancer.

---

Setup

Python environment

Requires Python 3.9+.

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

LINCS data files (required for LINCS step)

The LINCS L1000 knowledge graph files are too large for git. Download them from:

https://dd-kg-ui.cfde.cloud/downloads

Place the following files in the data/ directory:

• LINCS.Compound.nodes.csv
• LINCS.Gene.nodes.csv
• LINCS.edges.csv

The LINCS step can be skipped with --no-lincs if these files are unavailable.

---

Running the pipeline

python pipeline.py \
  --output-dir output \
  --search-terms "hepatocellular carcinoma" "HCC" \
  --gtex-tissue Liver \
  --gdc-project TCGA-LIHC

Key flags:

Flag	Default	Description
--output-dir	output	Directory for all result CSVs and JSON
--search-terms	HCC terms	BiomarkerKB condition search terms
--gtex-tissue	Liver	GTEx tissue (e.g. Pancreas, Lung, Colon_Transverse)
--gdc-project	TCGA-LIHC	TCGA project (e.g. TCGA-PAAD, TCGA-COAD)
--no-lincs	—	Skip LINCS L1000 step
--no-pharos	—	Skip Pharos/IDG step
--no-gdc	—	Skip GDC step
--no-string	—	Skip STRING PPI expansion
--cache-dir	cache	Local cache for API responses

Output files

Each run writes to --output-dir:

File	Contents
biomarkers.csv	BiomarkerKB hits
gtex_liver_expression.csv	Per-gene tissue TPM
pharos_targets.csv	IDG target annotations
gdc_stats.csv	Mutation freq + DE log2FC per gene
string_interactions.csv	PPI edges from STRING
lincs_perturbagen_hits.csv	Ranked perturbagens with annotations
final_scored_candidates_v2.csv	Final scored + ranked candidates
summary.json	Summary counts for each step

Scoring

After running the pipeline, score and rank candidates with:

python score.py --output-dir output --top 20

This reads lincs_perturbagen_hits.csv and writes final_scored_candidates_v2.csv with all component scores and a weighted total. Component score caps:

Component	Cap	Signal
BiomarkerKB	5	Number of biomarker genes hit
GTEx expression	15	Tissue TPM (log-normalized)
IDG/Pharos	8	Target development level
LINCS	10	Perturbagen signature strength
GDC/TCGA	8	Mutation freq + differential expression

---

Web dashboard

An interactive Django + React dashboard lets you explore results across all five cancers and adjust scoring weights in real time.

Start the backend

cd web
pip install -r requirements-web.txt
python manage.py migrate
python manage.py runserver

Start the frontend

cd web/frontend
npm install
npm run dev

Open http://localhost:5173 (or the Vite port shown in the terminal).

Features

• Disease selector — switch between HCC, Pancreatic, Colorectal, Lung, and Breast cancer results
• Summary cards — step-level counts (biomarkers, GTEx genes, targets, candidates)
• Scoring panel — ranked drug candidates with per-component score bars; five weight sliders (0–3×) re-rank in real time without re-running the pipeline
• BiomarkerKB, GTEx, Pharos, GDC, LINCS, STRING sections — paginated tables for each data source

The dashboard reads pre-computed CSV outputs. Run the pipeline at least once per disease before using the dashboard.

---

Repository layout

pipeline.py          # entry point: python pipeline.py [options]
score.py             # standalone scorer: python score.py [--output-dir ...]
src/
  pipeline.py        # pipeline orchestration
  scoring.py         # multi-component scoring module
  clients/           # API clients (BiomarkerKB, GTEx, Pharos, GDC, LINCS, STRING)
  models.py          # dataclasses for intermediate results
web/
  api/               # Django REST views + URL routing
  config/            # Django settings
  frontend/src/      # React components
data/                # LINCS files (not in git — see setup above)
output*/             # Pipeline results per disease (not in git)