PPI-Affinity: A Web Tool for the Prediction and Optimization of Protein–Peptide and Protein–Protein Binding Affinity

Abstract

Virtual screening of protein–protein and protein–peptide interactions is a challenging task that directly impacts the processes of hit identification and hit-to-lead optimization in drug design projects involving peptide-based pharmaceuticals. Although several screening tools designed to predict the binding affinity of protein–protein complexes have been proposed, methods specifically developed to predict protein–peptide binding affinity are comparatively scarce. Frequently, predictors trained to score the affinity of small molecules are used for peptides indistinctively, despite the larger complexity and heterogeneity of interactions rendered by peptide binders. To address this issue, we introduce PPI-Affinity, a tool that leverages support vector machine (SVM) predictors of binding affinity to screen datasets of protein–protein and protein–peptide complexes, as well as to generate and rank mutants of a given structure. The performance of the SVM models was assessed on four benchmark datasets, which include protein–protein and protein–peptide binding affinity data. In addition, we evaluated our model on a set of mutants of EPI-X4, an endogenous peptide inhibitor of the chemokine receptor CXCR4, and on complexes of the serine proteases HTRA1 and HTRA3 with peptides. PPI-Affinity is freely accessible at .

Detailed Molecular Interactions between Respiratory Syncytial Virus Fusion Protein and the TLR4/MD-2 Complex In Silico

“…The scoring functions can be roughly divided into four categories: force-field-based, empirical, knowledge-based, and machine-learning-based [38]. HDOCK, HawkDock, and PPI-Affinity use knowledge, force field, and machine learning-based scoring functions, respectively [24][25][26]. Thus, in addition to HDOCK, HawkDock and PPI-Affinity were used to determine the optimal docking model.…”

“…For the prefusion protein, the top 20 docking models were determined from the generated models at each antigenic site based on the docking score of HDOCK, and a total of 120 docking models were created. In addition to HDOCK, two independent scoring functions, HawkDock (http://cadd.zju.edu.cn/hawkdock/, accessed on 15 February 2022) and PPI-Affinity (https://protdcal.zmb.uni-due.de/PPIAffinity/BA/12 19/, accessed on 16 September 2022), were used to rescore the docking models generated by HDOCK [25,26]. The optimal models for the prefusion protein were selected based on the HDOCK, HawkDock, and PPI-Affinity ranking.…”

Machine learning methods for protein-protein binding affinity prediction in protein design

“…Machine learning methods, from simplest linear regression to deep learning (Seal, 1967;Cortes and Vapnik, 1995;Tin Kam Ho, 1995;Breiman, 1996;Friedman, 2002;LeCun et al, 2015;Goodfellow et al, 2016), have been developed for decades and are implemented in science, finance, healthcare, and other fields (Dixon et al, 2020;Guo et al, 2020;Varoquaux and Cheplygina, 2022;Zhang et al, 2022). In structural biology, machine learning methods have been used to predict the structure of proteins based on their amino acid sequences, design new molecules for enzyme inhibition, and predict the protein-protein interactions (Vamathevan et al, 2019;Baek et al, 2021;Jumper et al, 2021;Romero-Molina et al, 2022). In this section, we focus on regression models for protein-protein interaction prediction (Table 2).…”

“…The ensemble model achieved an R p of 0.853 on SKEMPI dataset. PPI-Affinity is a web tool that predicts the binding affinity using support vector machine and other classic machine learning models (Romero-Molina et al, 2022). Accepting thousands of features generated by ProtDCal as input (Romero-Molina et al, 2019), the model showed a performance of R p = 0.77 on SKEMPI dataset.…”

Towards rational computational peptide design

“…Finally, the prediction of structures and binding affinities seem to be on two independent tracks at the moment-some tools are good for sampling bound states and others for rank-ordering them by binding affinity (Cunningham et al, 2020;Chang and Perez, 2022a;Motmaen et al, 2022). While development in affinity prediction has lagged behind the structure prediction problem, there is a recent increase in the number of methods available, some of them now available as webservers with promising accuracy (Romero-Molina et al, 2022). It is feasible to think that as binding affinity databases increase in size and accuracy, these two independent pieces will be trained together.…”