International Conference on Computer and Knowledge Engineering

Home / 14th International Conference on Computer and Knowledge Engineering

Leveraging a structure-based and learning-based predictor using various feature groups in bioinformatics (case study: protein-peptide region residue-level interaction)

Authors :

Shima Shafiee¹ Abdolhossein Fathi²

1- Department of Computer Engineering and Information Technology Razi University Kermanshah, Iran 2- Department of Computer Engineering and Information Technology Razi University Kermanshah, Iran

Keywords :

Binding region residue،Deep learning،machine learning،protein-peptide interaction

Abstract :

Motivations: Predicting protein-peptide interactions is essential in cellular processes, researching protein function, new drug design, understanding abnormal cell behavior, and human diseases. Conventional experimental techniques for identifying protein-peptide interaction region residues are labor-intensive and expensive. Therefore, figuring out these binding region residues using computers would be a valuable and complementary tool. We introduce a computational method, RSPPRI (Residual neural network and Support vector machine-based prediction of Protein-Peptide Region residues-level Interaction), to detect protein-peptide binding region residues accurately. In this regard, various feature groups are extracted using protein structures, including evolutionary, and structure-based. Results: For two common test sets, the proposed method surpasses traditional and structure-based methods by an F-measure (F-M) of 0.326 with a sensitivity (SEN) of 64% and a specificity (SPE) of 68%. Importantly, our milestones show a 17.4% improvement in F-M and an improved balance (about 3%) between SEN and SPE scores. In addition, the proposed method successfully separates peptide-binding region residues from other functional region residues when applied to protein binding with various ligands, such as deoxyribonucleic acid, ribonucleic acid, and carbohydrates. Overall, the obtained results are robust and consistent for diverse binding region residue predictions. The findings demonstrate the proficiency of the proposed method.