EWCL–CAID Benchmarks 2025: Binding, Disorder (PDB & NOX), and Linker

Title: EWCL–CAID Benchmarks 2025: Independent Evaluation of Protein Disorder, Binding, and Linker Predictors Description: This dataset accompanies the 2025 EWCL (Entropy-Weighted Collapse Likelihood) CAID benchmark suite. It provides an independent, standardized evaluation of sequence- and structure-based predictors across four canonical CAID tasks: Binding-IDR, Disorder-PDB, Disorder-NOX, and Linker. The EWCL framework did not participate as a competing method in CAID; it served purely as an external benchmarking and reproducibility platform, applying the same protocols and thresholds used by CAID organizers to ensure compatibility and transparency. Each archive (EWCL_CAID_*.zip) includes: • Residue-level predictions and corresponding ground-truth annotations • Per-protein ROC-AUC and PR-AUC summary tables • Aggregate metrics (ROC-AUC, PR-AUC, F1-max, positive rate) • Visualization figures (ROC, PR, and score distributions in PNG format) Results were computed using EWCLv1, the structure-aware entropy-weighted model that integrates AlphaFold-derived features (pLDDT, entropy weighting, and curvature penalties). All files are aligned to CAID reference annotations for direct comparison.# 标题：EWCL–CAID 2025基准测试集：蛋白质无序区、结合位点与连接子预测器的独立评估 本数据集配套2025年EWCL（Entropy-Weighted Collapse Likelihood，熵权重折叠似然）CAID基准测试套件。 本数据集可针对四项标准CAID任务，对基于序列与结构的预测器开展独立、标准化评估，四项任务分别为结合-内在无序区（Binding-IDR）、无序-PDB、无序-NOX与连接子预测。 EWCL框架并未作为参赛方法参与CAID评测，仅作为外部基准测试与可复现性平台，采用CAID主办方使用的统一实验流程与阈值标准，以确保评测的兼容性与透明度。 每个归档文件（EWCL_CAID_*.zip）均包含以下内容： • 残基级预测结果及对应真实标注 • 单蛋白质受试者工作特征曲线下面积（ROC-AUC）与精确召回曲线下面积（PR-AUC）汇总表格 • 聚合评估指标（含ROC-AUC、PR-AUC、最大F1值、阳性率） • 可视化图表（含ROC曲线、PR曲线及得分分布的PNG格式文件） 评测结果基于EWCLv1计算得到，该模型为具备结构感知能力的熵权重模型，整合了阿尔法折叠（AlphaFold）衍生特征，包括pLDDT、熵权重与曲率惩罚项。 所有文件均与CAID参考标注对齐，可直接用于对比分析。