Machine Learning-Guided Identification of Ivermectin-Associated Biomarkers Across Human Cancers

This repository contains the computational pipeline used in the study:

Machine learning-guided identification of ivermectin-associated biomarkers reveals prognostic signatures and potential drug repurposing opportunities across human cancers

The project integrates transcriptomic data, supervised machine learning, explainable AI (SHAP), differential expression analysis, functional enrichment, Connectivity Map (CMap), and external validation using TCGA/GTEx to identify biomarkers associated with ivermectin sensitivity in human cancers.

---

Overview

Drug repurposing has emerged as an attractive strategy for accelerating cancer drug discovery. In this project, we investigated molecular signatures associated with ivermectin response using publicly available pharmacogenomic and transcriptomic datasets.

The workflow combines:

• transcriptomic profiling
• supervised machine learning
• SHAP explainable AI
• differential gene expression analysis
• GO, KEGG and Reactome enrichment
• Connectivity Map drug repurposing
• TCGA/GTEx validation
• survival analysis in TCGA-LIHC

The final machine learning dataset consisted of 220 cancer cell lines and 19,222 transcriptomic features.

---

Workflow

DepMap + PRISM
        │
        ▼
Data preprocessing
        │
        ▼
Machine learning
(Logistic Regression, Random Forest,
SVM, LightGBM, CatBoost)
        │
        ▼
SHAP feature importance
        │
        ▼
Differential expression
        │
        ▼
Functional enrichment
(GO / KEGG / Reactome)
        │
        ▼
Connectivity Map
        │
        ▼
TCGA / GTEx validation
        │
        ▼
Survival analysis (TCGA-LIHC)

The complete computational workflow is illustrated in Figure 1 of the manuscript.

---

Repository structure

.
├── data/              # Input datasets (not included)
├── docs/              # Documentation
├── figures/           # Figures generated during the study
├── manuscript/        # Manuscript
├── notebooks/         # Jupyter notebooks
├── results/           # Analysis outputs (optional)
├── scripts/           # Python scripts
├── tables/            # Result tables
├── environment.yml
└── README.md

---

Key Figures

Machine Learning Performance

ROC Curve	Confusion Matrix

---

Explainable AI

Top 25 features identified by SHAP analysis.

---

Differential Expression

Differentially expressed genes highlighted together with machine learning biomarkers.

---

Functional Enrichment

GO Biological Processes	KEGG Pathways

---

Transcriptomic Heatmap

---

Ivermectin Sensitivity Across Cancer Types

---

Machine Learning Models

Five supervised learning algorithms were evaluated:

• Logistic Regression
• Random Forest
• Support Vector Machine
• LightGBM
• CatBoost

Logistic Regression achieved the highest predictive performance and was selected for SHAP interpretation.

---

Main Findings

The integrated analysis identified eight candidate biomarkers:

• BAMBI
• ERBB3
• LOXL2
• TGFB2
• LAYN
• MYL9
• CAV1
• MIA

Functional enrichment highlighted pathways related to:

• extracellular matrix organization
• epithelial–mesenchymal transition (EMT)
• TGF-β signalling
• cell adhesion

Connectivity Map analysis suggested HDAC inhibitors as the most promising drug class for combination strategies with ivermectin.

---

Data Sources

• DepMap transcriptomic profiles
• DepMap PRISM drug sensitivity
• Connectivity Map (CMap)
• TCGA
• GTEx
• UCSC Xena

All datasets are publicly available.

---

Requirements

Python 3.12

Main libraries:

• pandas
• numpy
• scipy
• scikit-learn
• shap
• matplotlib
• seaborn
• statsmodels
• lifelines
• gseapy
• lightgbm
• catboost

Create the environment using

conda env create -f environment.yml
conda activate ivermectin-ml

---

Citation

If you use this repository, please cite:

Agata Gabara

---

License

MIT License

Author

Agata Gabara