BTEVAL

BTEVAL evaluates beta-turn prediction algorithms by benchmarking their amino-acid-level predictions against a standardized dataset of 426 non-homologous protein chains (≤25% sequence identity) to quantify predictive performance.

Key Features:

• Standardized dataset: A curated set of 426 non-homologous protein chains with no more than 25% sequence identity between any two chains and at least one beta turn per chain.
• Subset evaluation: Evaluation can be performed on the entire dataset or on specific subsets corresponding to seven protein groups.
• Amino-acid-level assessment: Predictions are assessed at the amino-acid level to determine per-residue correctness of beta-turn assignment.
• Performance metrics: Quantitative measures include Qtotal, Qpredicted, Qobserved, and Matthews Correlation Coefficient (MCC).
• Comparative benchmarking: Predictions are compared against established algorithms including the Chou-Fasman algorithm, Thornton's algorithm, GORBTURN, 1-4 and 2-3 Correlation model, Sequence coupled model, and BTPRED.

Scientific Applications:

• Benchmarking beta-turn predictors: Quantitatively compares new and existing beta-turn prediction algorithms against a uniform reference dataset using standard metrics.
• Method validation across protein groups: Assesses algorithm performance across seven protein-group subsets to reveal group-specific strengths and weaknesses.
• Support for protein-structure studies: Provides quantitative assessment of beta-turn prediction accuracy relevant to studies of protein folding and function.

Methodology:

Evaluate predictions on the 426-chain dataset or on any of seven protein-group subsets; assess at the amino-acid level; compute Qtotal, Qpredicted, Qobserved, and MCC; perform comparative analyses against Chou-Fasman, Thornton's algorithm, GORBTURN, 1-4 and 2-3 Correlation model, Sequence coupled model, and BTPRED.