DeepAIPs-SFLA: Deep Convolutional Model for Prediction of Anti-Inflammatory Peptides Using Binary Pattern Decomposition of Novel Multiview Descriptors with an SFLA Approach

Abstract

Inflammation is a vital biological response of the human immune system to harmful stimuli, and it plays a vital role in tissue repair and pathogen elimination. However, chronic inflammation can lead to severe diseases such as arthritis, cancer, cardiovascular disorders, and autoimmune conditions. Anti-inflammatory peptides (AIPs) have emerged as promising therapeutic agents owing to their high selectivity, potency toward target cells, and minimal side effects. Although numerous computational predictors exist for predicting AIP samples, most rely on traditional compositional features that fail to capture internal sequence ordering, local structural variations, and evolutionary information to determine peptide functionality. To address these problems, we propose DeepAIPs-SFLA, a novel deep learning-based computational model that integrates evolutionary information and structural features using advanced image-based encoding. The training sequences were transformed into two-dimensional structural and evolutionary images using RECM and PSSM embeddings. These images were further decomposed using LBP and CLBP algorithms, resulting in novel local texture descriptors: RECM_CLBP, PSSM_CLBP, and RECM_LBP. A differential evolution-based feature integration method was employed to construct a comprehensive multiview feature vector. Subsequently, an enhanced genetic algorithm-based shuffled frog-leaping algorithm (SFLA) was applied for optimal feature selection. An optimal feature set was used to train a deep residual convolutional neural network (RCNN). Our developed DeepAIPs-SFLA model attained an outstanding predictive accuracy of 97.04% with an AUC of 0.98 using the training sequences. The model was validated via independent sets to examine its generalization power, demonstrating substantial enhancements of 13 and 2% in accuracy compared with available predictors on the Ind-426 and Ind-1049 data sets, respectively. The robustness and efficacy of DeepAIPs-SFLA represent its potential as a valuable model for advancing academic research and drug discovery for inflammatory diseases.