iitanish/nppe2-protein-structure-data

该数据集包含了NPPE2蛋白质二级结构预测项目的预测结果和文档。数据集使用深度学习集成方法（BiLSTM、BiGRU、TransformerLSTM）进行蛋白质二级结构预测，并记录了关键的错误修复（词汇映射错误）带来的性能提升（+30%）。数据集包含测试集预测结果（1,816个序列）、技术报告（详细的方法论、错误修复文档和实验结果）和执行摘要（关键成果和主要结果）。预测格式包括8类和3类二级结构分类。性能指标显示测试F1得分为0.469，最佳验证F1得分为0.6287。数据集适用于学术和研究用途，采用MIT许可证。

This dataset contains the predictions and documentation from the NPPE2 Protein Secondary Structure Prediction project. It uses deep learning ensemble methods (BiLSTM, BiGRU, TransformerLSTM) for protein secondary structure prediction and documents a critical bug fix (vocabulary mapping error) that led to a 30% performance improvement. The dataset includes test set predictions (1,816 sequences), a technical report (detailed methodology, bug fix documentation, and experimental results), and an executive summary (key achievements and main results). The prediction format includes 8-class and 3-class secondary structure classifications. Performance metrics show a test F1 score of 0.469 and a best validation F1 score of 0.6287. The dataset is licensed under MIT for academic and research use.