Screening of oxidative stress components of cigarette smoke based on machine learning model integration

Abstract

Cigarette smoke, a complex mixture of more than 7000 chemicals, poses a significant threat to human health, with oxidative stress being an important mechanism in its associated diseases. Traditional methods for assessing the toxicity of cigarette smoke components, such as animal and cell-based assays, are often limited by their high cost and time consumption. This study integrates multiple machine learning algorithms and diverse data sources to construct a robust predictive model for identifying oxidative stress-inducing components in cigarette smoke. Utilizing a multi-dataset, multi-target and multi-algorithm modeling strategy, we developed an integrated model comprising 704 sub-models. These models were trained from 9 datasets related to reactive oxygen species (ROS)-associated pathways. The integrated model demonstrated better performance in external validation compared to individual models, predicting 974 ROS-positive components from 7111 cigarette smoke components. These components were clustered into 10 major classes, providing new insights into the structural diversity of oxidative stress-inducing components in cigarette smoke. Our findings offer a novel approach for enhancing the predictive capability of toxicity models and advancing the understanding of oxidative stress-related toxicity in cigarette smoke components.