BTEval: BTEVAL: A Server for Evaluation of β-Turn Prediction Methods

Welcome to the official repository for BTEVAL, a web server developed for assessing the performance of β-turn prediction methods and benchmarking them against existing algorithms on a standardized dataset. This resource is designed to support researchers in structural bioinformatics, protein structure prediction, and computational biology. Web Server: https://webs.iiitd.edu.in/raghava/bteval/ Citation Kaur, H., & Raghava, G. P. S. (2003). BTEVAL: A Server for Evaluation of β-Turn Prediction Methods. Journal of Bioinformatics and Computational Biology, 1(3), 495–504. https://doi.org/10.1142/S0219720003000253 Also cite: Kaur, H., & Raghava, G. P. S. (2002). An evaluation of β-turn prediction methods. Bioinformatics, 18(11), 1508–1514. About the Server BTEVAL is a benchmarking web server developed to address a long-standing challenge in β-turn prediction research: the lack of a uniform evaluation platform. Prior to BTEVAL, numerous β-turn prediction methods had been developed, but they were trained and tested on different datasets, making direct performance comparisons impossible. BTEVAL solves this by providing a standardized, clean dataset of 426 non-homologous proteins and a common set of performance metrics, allowing any newly developed β-turn prediction method to be rigorously evaluated and ranked against existing state-of-the-art methods. β-turns are critical non-repetitive structural elements in proteins that: Constitute on average 25% of all residues in protein chains Play key roles in protein folding, stability, and molecular recognition Have been the subject of numerous prediction methods over three decades Key Features Standardized Benchmark Dataset 426 non-homologous protein chains (BT426) No two chains share more than 25% sequence identity Each chain contains at least one β-turn β-turns assigned using the PROMOTIF program Seven subsets available for flexible cross-validation Comprehensive Performance Evaluation Evaluation at the amino acid level Four standard performance measures reported: Qtotal — overall prediction accuracy (% correctly classified residues) Qpredicted — probability of correct prediction (precision) Qobserved — percentage coverage of observed β-turns (recall/sensitivity) MCC — Matthews Correlation Coefficient (accounts for over- and under-prediction) ROC curve generation with and without cross-validation