# Network toxicology reveals key genes of amiodarone induced pulmonary fibrosis: based on machine learning and SHAP analysis

## TL;DR

This study identifies key genes involved in amiodarone-induced pulmonary fibrosis using machine learning and experimental validation, offering new insights into the condition's causes and potential treatments.

## Contribution

The novel contribution is the identification of CTSK, ADORA3, and AGER as key genes in amiodarone-induced pulmonary fibrosis using machine learning and experimental validation.

## Key findings

- CTSK, ADORA3, and AGER were identified as the most important predictors of AIPF using SHAP analysis.
- AMD treatment altered the expression of CTSK, ADORA3, and AGER in bronchial epithelial cells.
- Molecular simulations confirmed stable binding between AMD and the identified target proteins.

## Abstract

Amiodarone (AMD), a highly effective Class III antiarrhythmic drug, has its clinical utility limited by the risk of inducing a serious adverse effect, amiodarone-induced pulmonary fibrosis (AIPF). The pathogenesis of AIPF remains poorly elucidated, particularly the hub driver genes, which hinders early diagnosis and targeted intervention.

This study employed an integrative approach combining network toxicology, machine learning (ML), and in vitro validation to identify hub genes in AIPF. Potential AMD targets and pulmonary fibrosis (PF)-related genes were obtained from toxicity databases and transcriptomic data (GEO datasets), respectively, and intersected to identify candidate AIPF targets. Multiple ML models were constructed, and SHAP (Shapley Additive exPlanations) analysis was used to interpret the model and rank feature importance. Molecular docking and dynamics simulations assessed the binding of AMD to the core targets. Key findings were experimentally validated in an AMD-induced human bronchial epithelial (BEAS-2B) cell model using qRT-PCR, Western blot, and functional assays.

Bioinformatics analysis identified eight candidate hub genes for AIPF. The glmBoost + GBM model demonstrated superior predictive performance (AUC = 0.845). SHAP interpretability analysis identified Cathepsin K (CTSK), Adenosine A3 Receptor (ADORA3), and Advanced Glycosylation End Product-Specific Receptor (AGER) as the most important predictors. Molecular simulations confirmed stable binding between AMD and these target proteins. In vitro experiments showed that AMD treatment significantly upregulated CTSK and downregulated ADORA3 and AGER at both mRNA and protein levels in BEAS-2B cells, and enhanced cell migration and invasion.

This study identifies CTSK, ADORA3, and AGER as key genes in AIPF pathogenesis through a comprehensive bioinformatics and ML approach. Their dysregulation in lung epithelial cells likely promotes fibrosis through modulating extracellular matrix metabolism, inflammation, and cell motility. These findings provide novel insights into AIPF mechanisms and highlight potential biomarkers and therapeutic targets.