Protein-Peptide Interactions in Regulatory Events

Raghavender, Upadhyayula S.; Rathore, R. S.

doi:10.1016/b978-0-12-809633-8.90694-0

Cited by 1 publication

(1 citation statement)

References 62 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Three major categories of ML approaches exist to classify a given system: unsupervised, supervised, and reinforcement learning. A very few efforts have been proposed in the last few decades to model PPepIs using machinelearning or deep-learning methods (7)(8)(9)(10). We have performed supervised machine learning algorithms like Bayesian Network (Bayes Net), Random Forest, Logistic Regression, Adaptive Boosting (AdaBoost), Arti cial Neural Network (ANN), and Simple Logistic methods to classify PPIs and PPepIs using three different datasets of 636 PPIs, PPepIs, and non-interacting complexes, consisting of 56, 73, and 9 features.…”

Section: Introductionmentioning

confidence: 99%

Discrimination of Protein-Protein and Protein-Peptide Interactions using Machine Learning Methods

Kumar,

Rathore

2024

Preprint

View full text Add to dashboard Cite

Protein-protein interactions (PPI) play important roles in almost all cellular processes. PPI also includes protein-peptide interactions (PPepI), which, by an estimate, account for 15–40% of all such interactions. Even though protein-protein and protein-peptide recognition mechanisms sound similar, seemingly subtle differences exist among them. Knowledge of such differences is essential for biologics design when augmentation or disruption of protein-protein interactions is substituted with peptide-based mimics. Peptide-based leads have multiple advantages, including longer shelf life, feasibility of oral delivery, flexibility of optimisation, screening, and versatility of mimetics synthesis. To characterise differences between protein-protein and protein-peptide interactions, we have used machine learning approaches to classify these interactions. We compiled three datasets, comprising protein-protein, protein-peptide, and non-interacting protein complexes, each of which has 212 high-quality crystal structures. We calculated 583 sequence and physicochemical properties based on the on the features of one protein partner in all three datasets. With the correlation-based feature selection attribute evaluator and the best first search method, 56 features were chosen for classification. We performed different supervised machine learning algorithms with a 10-fold cross-validation method for unbiased classification of PPI and PPepI datasets. The Bayesian network method yielded the highest accuracy of 80.53%. Further, to achieve the highest accuracy and identify relevant features that can be employed for peptide-based biologic design applications, we calculated 73 PPI-specific molecular descriptors and applied ML methods. Application of the simple logistic method resulted in the highest accuracy of 92.21% (recall 0.925, precision 0.920, ROC 0.978, and PRC 0.981). The present attempt to discriminate between the between the recognition mechanisms of protein-protein and protein-peptide in cellular processes should aid in the design of peptide-based biologics.

show abstract

Section: Introductionmentioning

confidence: 99%