Year: 2025
Type: article
Abstract: Antigenicity prediction plays a crucial role in vaccine development, antibody-based therapies, and diagnostic assays. The methodologies to perform antigenicity prediction are commonly applied to entire protein sequences and epitopes discovered through reverse vaccinology. These methods help to assess the potential of these structures to induce and recruit human defense cells and the development of antibodies. While numerous methods have been proposed for antigenicity prediction, they often come with specific constraints linked to the input sequence origin and feature extraction techniques, with predictions more commonly focused on viruses and immunotherapy. This 1 study presents the Pipeline for Antigenicity Prediction Comparison (PAPreC), a versatile tool for exploring key factors in antigenicity prediction from bacterial epitopes. It offers various prediction options and supports personalized model development, demonstrating its applicability beyond bacteria to other organisms. PAPreC systematically explores three key factors: the source dataset used for training (including short peptides and complete protein sequences), the feature extraction methods employed (utilizing covariance matrices and physicochemical descriptors), and a diverse set of experimental configurations encompassing different classifiers. By analyzing the impacts of unbalanced data, classifier selection, and training dataset choices, PAPreC aims to provide a wide range of prediction options for the proteins and epitopes while offering a personalized model from feature selection to refine and turn the epitope curation efficiently. Besides, we present a case study for two clinical pathogenic strains of species Pseudomonas aeruginosa and Staphylococcus aureus, whose genomes were previously sequenced by our group. The findings of our study demonstrate that specific feature configurations exhibit more outstanding suitability for different types of protein sequences. Furthermore, based on the bacterial study, the methodologies utilized in this research eliminate the need for separate models explicitly trained for different gram types (gram-positive and gram-negative) of bacterial sequences.
Source: ACS Omega
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $1,935
Funders: Universidad Nacional de Córdoba, Coordenação de Aperfeiçoamento de Pessoal de Nível Superior, Consejo Nacional de Investigaciones Científicas y Técnicas, Conselho Nacional de Desenvolvimento Científico e Tecnológico, Fundação Carlos Chagas Filho de Amparo à Pesquisa do Estado do Rio de Janeiro, Agencia Nacional de Promoción Científica y Tecnológica
OpenAlex ID: W4407113256