Abstract
Accurate prediction of interleukin-13 (IL-13)-inducing epitopes is crucial for advancing targeted therapies against allergic inflammation, the cytokine storm associated with severe COVID-19, and related disorders. Current epitope prediction methods, however, often exhibit limitations in efficiency and accuracy. To address this, we introduce DeepEpilL13, a novel deep learning framework that uniquely synergizes pretrained language models with multiwindow convolutional neural networks (CNNs) for the rapid and accurate identification of IL-13-inducing epitopes from protein sequences. DeepEpilL13 leverages high-dimensional embeddings generated by the pretrained language model, which capture rich contextual information from protein sequences. These embeddings are then processed by a multiwindow CNN architecture, enabling the effective exploration of both local and global sequence patterns pertinent to IL-13 induction. The proposed DeepEpilL13 approach underwent rigorous evaluation using both benchmark data sets and an independent SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) data set. Results demonstrate that DeepEpilL13 achieves superior performance compared with traditional methods. On the benchmark data set, DeepEpilL13 attained a Matthews correlation coefficient (MCC) of 0.52 and an area under the receiver operating characteristic curve (AUC) of 0.86. Notably, when assessed on the independent SARS-CoV-2 data set, DeepEpilL13 exhibited remarkable robustness, achieving an MCC of 0.63 and an AUC of 0.92. These metrics underscore the enhanced predictive capability and robust applicability of DeepEpilL13, particularly within the context of the COVID-19 research and related viral infections. This study presents DeepEpilL13 as a powerful and efficient deep learning framework for accurate epitope prediction. By offering significant improvement in performance and robustness, DeepEpilL13 provides new and promising avenues for the development of epitope-based vaccines and immunotherapies specifically targeting IL-13-mediated disorders. The successful and rapid identification of IL-13-inducing epitopes using DeepEpilL13 paves the way for novel therapeutic interventions against a range of conditions, including allergic diseases, inflammatory conditions, and severe viral infections such as COVID-19, with potential for a significant impact on public health outcomes.