iPIPs-sABiTCN: identifying proinflammatory peptides using local phase quantization based localized descriptors with self-attention Bidirectional Temporal Convolutional Network

Proinflammatory peptides (PIPs) play a significant role in regulating inflammatory responses and are intricately linked to the progression of several inflammatory disorders. Although current experimental and machine learning-based methods for predicting PIPs have achieved significant results, they are labor-intensive, expensive, and unable to provide deeper contextual and structural representations of peptide samples. In this study, we present a novel computational predictor, iPIPs-sABiTCN, for the accurate prediction of PIPs. The input peptides were represented in the form of a 2D matrix using a position-specific scoring matrix (PSSM) and substitution matrix representation (SMR). Subsequently, the generated 2D matrix is passed through local phase quantization (LPQ) to produce the intrinsic local and evolutionary structure-based descriptors, namely, SMR-LPQ and LPQ-PSSM. Multiple variants of the evolutionary scaling matrix (ESM-2) were investigated for contextual representation. Additionally, a BTGA + KNN-based genetic algorithm is utilized to select the high-ranked features from the integrated hybrid vector. Among several classifiers, the proposed self-attention-based Bidirectional Temporal Convolutional Network (sABiTCN) model demonstrated superior predictive performance. The proposed iPIPs-sABiTCN predictor achieved the highest training accuracy of 87.30% with an AUC of 0.94, outperforming the predictive accuracy of ~ 18% and AUC of ~ 23% of existing PIP models. Further validation revealed that the iPIPs-sABiTCN training model outperformed two unseen independent datasets, demonstrating improvements of approximately 15% and 12% on the Ind-Set1 and Ind-Set2 test sets, respectively. The generalized efficiency and consistency of the iPIPs-sABiTCN model highlight its potential as a valuable tool in academic research, diagnosis, and treatment of different inflammatory diseases. The user-friendly web server of our proposed predictor is publicly accessible at https://ipipssabitcn.pythonanywhere.com/ .