Abstract:
The original CBTOPE method, introduced in 2009, was one of the first approaches for predicting conformational B-cell epitopes using only protein sequence information. While it has been widely adopted in the scientific community, recent benchmarking by Cia et al. (2023) revealed that many existing tools, including CBTOPE, exhibit suboptimal performance on rigorously curated and expanded datasets. This performance gap is largely attributed to limitations in earlier training data and feature representation. In response, we present CBTOPE2, an improved version of the original model, developed using modern machine learning techniques and trained on the high-quality dataset curated by Cia et al. (2023). The enhanced pipeline integrates multiple types of features, including binary profile, position-specific scoring matrices (PSSM), and relative solvent accessibility (RSA), to better capture evolutionary and structural signals relevant to epitope recognition. Initial models using binary features achieved a maximum AUC of 0.61 on the validation set. The addition of evolutionary features via PSSM improved performance to 0.67, while combining PSSM with RSA further boosted the AUC to 0.70. Multiple classifiers were evaluated, with Gradient Boosting trained on PSSM+RSA features yielding the best results: AUC = 0.70 and MCC = 0.28 on an independent validation set. To support practical application, CBTOPE2 is available as both a web server and a standalone Python package. The web platform allows users to submit antigen sequences in FASTA format and receive predictions of antibody-binding residues, while the standalone version (pip install cbtope2) supports offline analysis. The CBTOPE2 platform (https://webs.iiitd.edu.in/raghava/cbtope2) provides a significantly improved and accessible tool for conformational B-cell epitope prediction, designed to aid researchers in vaccine development, antibody engineering, and immunodiagnostics.
