Abstract
Therapeutic proteins are in high demand due to their significant potential, driving continuous market growth. However, a critical concern for therapeutic proteins is their ability to trigger an immune response, while some treatments rely on this response for their therapeutic effect. Therefore, to assess the efficacy and safety of the drug, it is pivotal to determine its immunogenicity potential. Various experimental methods, such as cytokine release or T-cell proliferation assays, are used for this purpose. However, these assays can be costly, time-consuming, and often limited in their ability to screen large peptide sets across diverse major histocompatibility complex (MHC) alleles. Hence, this study aimed to develop a computational classification model for predicting the release of interferon-gamma based on the peptide sequence and the MHC class II (MHC-II) allele pseudo-sequence, which represents the binding environment of the MHC-II molecule. The dataset used in this study was obtained from the Immune Epitope Database and labeled as active or inactive. Among the approaches explored, the random forest algorithm combined with letter-based encoding resulted in the overall best-performing model. Consequently, this model's generalizability to other T-cell activities was further evaluated using a T-cell proliferation dataset. Furthermore, feature importance analysis and virtual single-point mutations were conducted to gain insights into the model's decision-making and to improve the interpretability of the model.