Machine Learning-Based Single-Cell RNA-seq Analysis in Drug Response: A Hybrid Multi-Stage Gene Selection and Soft Voting Approach

Khalsan, Mahmood; Al-Khafaji, Alia H.; Al-Alloosh, Fawaz

doi:10.31557/APJCP.2026.27.10.3709

Machine Learning-Based Single-Cell RNA-seq Analysis in Drug Response: A Hybrid Multi-Stage Gene Selection and Soft Voting Approach

Document Type : Research Articles

Authors

¹ Scientific Department, Warith International Cancer Institute, Karbala, 56001, Iraq.

² College of Science, Information Technology, The University of Warith Al-anbiyaa, Karbala, 56001, Iraq.

10.31557/APJCP.2026.27.10.3709

Abstract

Objective: Oral squamous cell carcinoma (OSCC) is the most common type of oral cancer, and cisplatin is widely used as a first-line chemotherapeutic agent. However, resistance to chemotherapy remains a significant challenge in the treatment of OSCC. The development of chemoresistance is driven by various molecular mechanisms and is further complicated by interactions within the tumor microenvironment (TME). Recent advances in high-throughput sequencing technologies, such as single-cell RNA sequencing (scRNA-seq), have enable the analysis of gene expression at the single-cell level. This approach provides deeper insights into complex intratumoral heterogeneity and facilitates the development of more targeted and effective therapies. Nevertheless, the large volume of data generated by scRNA-seq presents challenges for traditional statistical methods, a limitation that has been effectively addressed through advancements in machine learning. Methods: In this study, we developed a hybrid multi-stage gene selection framework combined with a soft voting classifier to predict cancer drug response at the single-cell level by integrating large-scale bulk cell line data. A total of 100 significant genes were selected as features for training the developed soft voting classifier. Result: The results demonstrated the effectiveness of the proposed machine learning model, achieving an accuracy of 97.9%, precision of 98.9%, recall of 96.9%, and an F1-score of 97.9% in predicting cisplatin response in patient-derived OSCC Cells at the single-cell level. Conclusion: This study highlights the effectiveness of combining single-cell RNA sequencing with machine learning to predict cisplatin response in oral squamous cell carcinoma. The proposed hybrid multi-stage gene selection and soft voting classifier successfully captured intratumoral heterogeneity at the single-cell level. Using 100 selected genes, the model achieved high predictive performance, demonstrating strong accuracy and robustness. These findings emphasize the value of machine learning in analyzing high-dimensional scRNA-seq data. Overall, this approach provides a promising foundation for precision medicine and improved therapeutic decision-making in OSCC.

Keywords