AIPFusion-Net: an interaction-fusion framework for anti-inflammatory peptide prediction using protein language model embeddings, local sequence patterns, and handcrafted descriptors

Introduction Anti-inflammatory peptides (AIPs) are promising candidates for treating inflammatory diseases. However, experimental identification is time-consuming and costly, creating a need for accurate computational prediction methods. Methods We developed AIPFusion-Net, an interaction-fusion framework that integrates three complementary sources of peptide information: global contextual embeddings generated by the ESM2 protein language model, local sequence patterns extracted from biologically relevant subsequences, and handcrafted physicochemical and sequence descriptors. These representations were integrated through an interaction-fusion module and evaluated on two independent test datasets, Ind-1049 and Ind-426. Results On the Ind-1049 dataset, AIPFusion-Net achieved an accuracy of 84.6%, sensitivity of 71.7%, specificity of 89.3%, MCC of 0.654, and AUC of 0.884. On the Ind-426 dataset, it achieved an accuracy of 82.3%, sensitivity of 84.4%, specificity of 76.9%, MCC of 0.673, and AUC of 0.882. The model outperformed the comparative predictors in overall classification performance and discriminative ability. Discussion The results indicate that combining global contextual information, local sequence patterns, and handcrafted descriptors through interaction fusion improves the representation and identification of AIPs. AIPFusion-Net provides a useful computational framework for AIP screening and may support the discovery and design of novel anti-inflammatory peptides.