Prediction of Interleukin-binding Sites Combining Multi-Source Features with Integrated Algorithm

Introduction: Interleukins (ILs) are important immune cytokines involved in immune regulation, inflammatory responses, and metabolic control. They are closely associated with various diseases, such as rheumatoid arthritis, atherosclerosis, diabetes, and asthma. However, the specific binding mechanisms of interleukins remain unclear. Studying the binding mechanisms between proteins and interleukins can help to understand the functions of interleukins, disease pathogenesis, and the development of new drugs. This study aims to systematically analyze the characteristics of interleukin family binding sites, uncover their shared features and specific mechanisms, provide new perspectives for understanding their functional roles in ligand-receptor interactions, and elucidate the potential impact of binding sites on signal transduction and immune responses. Methods: We constructed a dataset containing both binding and non-binding sites. Extracted eight features based on the sequence, structure, and functional information of the proteins. Six machine learning algorithms, along with an integrated algorithm, were used to predict these features. Results: We found that among the machine learning algorithms, the prediction performance using energy features was the best, achieving the highest accuracy (ACC) and area under the ROC curve (AUC). Further feature fusion and ensemble algorithm models significantly improved the predictive performance, with a maximum accuracy (ACC) of 98.4% and an ensemble algorithm accuracy of up to 99.2%. Discussion: This study outperforms existing methods, achieving an MCC score of 0.984 with the Gradient Boosting algorithm. However, the limitations of a small sample size and dataset imbalance highlight the need for future research to collect larger and more diverse datasets to improve the model';s generalization ability and predictive accuracy. Future studies will aim to verify our method';s applicability and develop an online prediction tool to assist in studying small molecule drugs, antibodies, and interleukin binding sites, supporting targeted drug design and treatment of immunerelated diseases. Conclusion: This study demonstrates that the developed predictive model for interleukin binding sites effectively utilizes geometric and biochemical features, validating the SMOTETomek sampling method in enhancing model performance and providing a basis for targeted drug design and understanding immune response mechanisms.