Integrative transcriptome-based drug repurposing in tuberculosis

This repository contains the analysis code and processed data for the study:

Samart K, Thang L, Buskirk L, Tonielli A, Krishnan A*, Ravi J*. Integrative transcriptome-based drug repurposing in tuberculosis. bioRxiv (2025). https://doi.org/10.1101/2025.06.02.657296

* co-corresponding authors

We integrate transcriptomic signatures from multiple TB microarray and RNAseq datasets with various disease-drug connectivity scores to identify and prioritize HDT candidates for tuberculosis.

---

Repository structure

integrative-drugrep-tb/
├── scripts/         # R and Python analysis scripts (numbered by step)
├── vignette/        # Quarto documents with worked examples for each step
├── evaluation/      # Pubmed scraper and outputs
├── figures/         # Code and outputs for manuscript figures (figure1–6, figureS1-S16)
├── data/            # Input data: DE results, signatures, and metadata
├── DataCuration/    # Notebooks for preprocessing, cleaning, and harmonizing disease data and metadata
├── results/         # Pipeline outputs: connectivity scores and drug rankings
└── renv.lock        # R package lockfile for reproducibility

---

Workflow

The analysis follows a five-step pipeline:

Step	Scripts	Description
00	00_multids_microarray_DEwithlimma.R, 00_multids_RNAseq_DEwithDESeq2.R	Differential expression analysis across TB datasets
01	01_signature_aggregation_functions.R, 01_signature_landmark_prep_functions.R	Aggregate DE signatures; filter to L1000 landmark genes
02	02_drugrep_get_prediction_indiv.R, 02_drugrep_get_prediction_aggr.R, 02_signatureSearch_connectivity_scores_functions.R	Score drug candidates via CMAP 1.0, CMAP 2.0 (LINCS), and correlation methods
03	03_summarize_drugs_methodswise_functions.R	Summarize and compare predictions across scoring methods
04	04_*_RankAggregation_*.R	Aggregate drug rankings across methods and datasets
05	05_high_confidence_drug_prediction.R, 05_build_drugtarget_network_functions.R	Identify high-confidence candidates and build drug–target networks

See scripts/README.md for a full table of scripts and their purposes, and vignette/ for step-by-step worked examples.

---

Reproducing the analysis

Requirements

• R 4.6.1 with Bioconductor 3.23
• Python 3 (for HDT literature evidence retreival and baseline analysis

Setup

# Install renv if needed
install.packages("renv")

# Restore the R package library
renv::restore()

Running the pipeline

1. Generate and aggregate individual disease signatures

1.1 Microarray individual disease signatures

Rscript scripts/00_multids_microarray_DEwithlimma.R \
  data/microarray_data_forDE/clean_TB_sample_metadata_classification.tsv \
  0.05

Arguments:

• meta_class_file.tsv : Tab‑separated file listing all datasets (default data/microarray_data_forDE/clean_TB_sample_metadata_classification.tsv).

  Mandatory columns:

  • series_id (GEO study identifier)
  • geo_accession (GEO sample identifier )
  • signature_full_name (name of signature containing unique sample conditions)
  • exprmat_path (path to raw‑count matrix TSV)
  • classification (labels: disease_without_treatment or healthy_control_without_treatment)

• padj_cutoff : Adjusted‑p significance threshold (default 0.05)

1.2 RNAseq individual disease signatures

Rscript scripts/00_multids_RNAseq_DESeq2.R \
  data/RNAseq_data_forDE/clean_TB_sample_metadata_classification.tsv \
  0.05

Arguments:

• meta_class_file.tsv : Tab‑separated file listing all datasets (default data/RNAseq_data_forDE/clean_TB_sample_metadata_classification.tsv).

  Mandatory columns:

  • series_id (GEO study identifier)
  • geo_accession (GEO sample identifier )
  • signature_full_name (name of signature containing unique sample conditions)
  • exprmat_path (path to raw‑count matrix TSV)
  • classification (labels: disease_without_treatment or healthy_control_without_treatment)

• padj_cutoff : Adjusted‑p significance threshold (default 0.05)

1.3 Compute aggregated signatures

Aggregated signatures can be computed by running the Quarto (.qmd) notebook below.

vignette/01_compute_aggregated_signatures.qmd

2. Prioritize drug candidates using multiple connectivity scores

Drug candidates are prioritized by computing connectivity scores between disease signatures and drug perturbation signatures.

2.1 Get drug predictions for individual disease signatures

An example command to quantify candidate drugs predicted to reverse RNAseq individual TB signatures using the CMAP 2.0 methods (i.e., LINCS).

Rscript scripts/02_drugrep_get_prediction_indiv.R \
  data/signatures/RNASeq_TB_signature_run_info.tsv \
  data/signatures/RNAseq \
  LINCS LINCS \
  results/RNAseq/LINCS

Arguments:

• sig_metadata_path Path to the RNAseq signature metadata file (default: data/signatures/RNASeq_TB_signature_run_info.tsv).
• sig_data_path Directory containing RNAseq signature data files (default: data/signatures/RNAseq).
• drugdb_name Drug perturbation database to use. Options: LINCS, CMAP (default: LINCS).
• score_method Method used to compute signature similarity scores. Options: LINCS, CMAP, Cor_spearman, Cor_pearson (default: LINCS).
• output_dir Directory where output results will be saved (default: results/RNAseq/LINCS).

2.2 Get drug predictions for aggregated disease signatures

An example command to quantify candidate drugs predicted to reverse RNAseq aggregated TB signatures using the CMAP 2.0 methods (i.e., LINCS).

Rscript scripts/02_drugrep_get_prediction_aggr.R RNAseq LINCS LINCS

Arguments:

• technology Type of transcriptomic data used to generate aggregated signatures. Options: microarray or RNAseq (default: RNASeq)
• score_method Method used to compute disease-drug connectivity scores. Options: LINCS, CMAP, Cor_spearman, Cor_pearson (default: LINCS).
• drugdb_name Drug perturbation database to use for querying signatures. Options: LINCS, CMAP (default: LINCS).

Note: Drug prioritization examples by method for both individual and aggregated signatures can be reproduced by running the following notebook.

vignette/02_signatureSearch_connectivity_score_functions.qmd

3. Summarize drug predictions

Drug predictions from different connectivity scoring methods are aggregated and summarized using the following analysis notebooks:

3.1 Prioritize predicted drug candidates based on their consistency of reversal across methods

vignette/03_summarize_drugs_methodswise_functions.qmd

3.2 Integrate drug rankings across multiple methods using partial rank aggregation

vignette/04_partial_rank_aggregation.qmd

3.3 Identify high-confidence drug candidates based on aggregated rankings

vignette/05_high_confidence_drug_prediction.qmd

NOTES:

• Additional downstream analyses (e.g., pathway and baseline analyses) can be reproduced using the .qmd notebooks provided in the vignettes/ directory.
• All figures can be reproduced using the .qmd notebooks provided in the figures/ directory.

---

Citation

If you use this code or data, please cite:

Samart K, Thang L, Buskirk L, Tonielli A, Krishnan A*, Ravi J*.
Integrative transcriptome-based drug repurposing in tuberculosis.
bioRxiv (2025). https://doi.org/10.1101/2025.06.02.657296
(* co-corresponding authors)

A CITATION.cff file is also provided for automated citation tools.

---

License

MIT © 2026 JRaviLab & KrishnanLab. See LICENSE.

AI disclosure

Claude Code (model: Sonnet) was used to check for filename and other minor inconsistencies across the repository; any other use of LLMs for generating code skeletons is cited within each script.