Drug to Drug Interaction Multiclass Prediction on Biomedical Literature Knowledge Graphs

Abstract

Lung cancer remains a major global health concern, requiring continued advances in therapeutic approaches. The complex landscape of lung cancer therapies introduces the potential for complex drug-drug interactions (DDIs), affecting patient outcomes and treatment efficacy. The conventional study of DDIs, based on experimental methods, is limited by time, cost and scope. Recent advances in graph databases and machine learning provide an opportunity to accelerate and enhance DDI prediction.

This MSc thesis explores the fusion of graph databases and machine learning to predict drug interactions in lung cancer. By leveraging the interconnection of graph databases, the relationships between lung cancer drugs, target proteins, pathways, and side effects are encapsulated, enabling a structured and scalable representation of complex interactions.

The central goal is to devise an accurate machine learning framework for DDI prediction among lung cancer drugs. Analyzing the structure of the graph database facilitates the extraction of key features and patterns that facilitate interaction prediction, including competitive effects. A comprehensive knowledge graph is constructed, which contains various aspects of lung cancer drugs and their interactions. Machine learning algorithms are used to build predictive models capable of understanding complex DDI patterns, and the results of this research contribute to personalized medicine and deepen the understanding of lung cancer treatments.

The following chapters present the methodology, including data acquisition, modeling techniques, and experimental results. By evaluating the effectiveness and feasibility of using graph databases and machine learning in predicting drug interactions for lung cancer, this study seeks to improve the landscape of drug interaction prediction and promote more effective treatments for lung cancer.