Machine learning-based identification of exosome-related biomarkers and drugs prediction in nasopharyngeal carcinoma

Exosomes are recognized as essential mediators in the intercellular communication between tumor cells, serving a pivotal function in tumor development. Nevertheless, the patterns of expression and medical relevance of exosome-related genes (ERGs) in nasopharyngeal carcinoma (NPC) remain insufficiently characterized.

Datasets retrieved from the Gene Expression Omnibus database were consolidated into a comprehensive gene dataset, which was then employed to ascertain differentially expressed genes (DEGs) by comparing NPC samples with controls. ERGs were intersected with the DEGs, yielding the detection of exosome-related DEGs. These identified genes underwent functional annotation and pathway enrichment evaluation. The least absolute shrinkage and selection operator regression, support vector machine, and random forest approaches were utilized to develop NPC diagnostic model. Key genes were determined through intersection analysis and subsequently confirmed in an independent cohort. Furthermore, drug screening, molecular docking, and molecular dynamics simulation were executed to generate meaningful insights for developing therapeutic compounds.

Through the application of three machine learning algorithms, five key genes (LTF, IDH1, ITGAV, CCL2, and LGALS3BP) were identified for the construction of a diagnostic model. Validation results demonstrated the strong discriminative and calibration abilities of the model. Furthermore, molecular docking analysis revealed that the interaction between IDH1 and nelfinavir exhibited the lowest Vina score, suggesting a stable binding affinity.

This study identifies five exosome-related key genes, utilizing machine learning approaches to develop a diagnostic model and uncover potential drug targets for NPC. These findings offer novel perspectives for both the diagnosis and therapeutic development of NPC.