A comprehensive review and evaluation of machine learning-based approaches for identifying tumor T cell antigens

The precise identification of tumor T-cell antigens (TTCAs) is crucial for advancements in cancer immunotherapy and other clinical uses. In contrast to the labor-intensive and time-consuming process of experimentally identifying TTCAs, computational prediction offers a complementary approach by providing a shortlist of probable TTCA candidates for further experimental validation. Currently, several computational approaches, primarily based on machine learning (ML) methods, have garnered considerable attention for the in silico identification of tumor T-cell antigens (TTCAs). Therefore, this study presents a comprehensive survey on the existing state-of-the-art TTCA predictors. Based on our research, this is the first comprehensive review focused on both traditional ML and ensemble learning methods for TTCA identification. Specifically, we examine critical aspects of TTCA predictor development, including core algorithms, methodologies, benchmark datasets, feature encoding methods, feature selection approaches, and web server usability. We then analyze and compare the effectiveness and robustness of existing predictors across well-known benchmark datasets and case studies. Finally, we provide a detailed summary of the advantages and disadvantages of current TTCA predictors, along with essential insights and suggestions for developing novel computational approaches to accurately identify TTCAs. The insights gained from this review and benchmarking survey are expected to offer valuable guidance to researchers, aiding in the development of high-accuracy TTCA predictors for improved antigen identification in the future.