Forecast of cytotoxic T lymphocyte epitope using sequence weighting and artificial neural network based on EasyPred modeler

Aim: Cytotoxic T lymphocytes (CTL) examine the major histocompatibility complex (MHC) class I ligands on nucleated cells to detect antigens derived from pathogens and cancer cells. Accurate prediction of T-cell epitopes is therefore crucial for the development of a wide range of biopharmaceuticals, including vaccines. Methods: The present study involved the development of position-specific scoring matrices (PSSM) and artificial neural networks (ANN) based models for 22 MHC class I molecules, including the integrated forecast of CTL epitopes using the EasyPred modeler. Similarity-reduced peptides dataset was used to train and evaluate models with performance assessed using the area under the receiver operating characteristic curve (Aroc) as the primary metric. Results: Comparative analysis revealed that the ANN-based predictor achieved superior performance for the HLA-A*0202 molecule by achieving the maximum Aroc value of 0.97 as compared to the PSSM predictor, having a value of 0.93. Furthermore, most natural MHC binders were identified within the top 5% with an average relative rank (%) of 2.23 and 3.13 for predictors PSSM and ANN, respectively, on the NetCTLpan dataset. Likewise, evaluation on the SARS-CoV-2 dataset of HLA-A*0201 revealed that the PSSM predictor (2.46%) performed better than the other contemporary CTL epitope forecast methods like naturally eluted ligands (EL) of NetMHCpan 4.0 (2.66%), NetCTLpan 1.1 (2.69%), and binding affinity (BA) of NetMHCpan 4.0 (3.33%), respectively. Conclusions: The application of these predictive models offers a significant reduction of approximately 97% in the resources typically required for epitope identification, including costs related to materials, labor, and time. As such, these models represent a valuable advancement in the rational design of more efficient, cost-effective, and innovative biotherapeutics.