multimolecule/ipknot_plus_plus
收藏资源简介:
IPknot++是一个用于RNA二级结构预测的基准数据集,随IPknot++论文发布。该数据集旨在作为外部评估集,而非训练分割,通常用于在SPOT-RNA风格的分割(如bpRNA-spot)上训练或调优RNA二级结构预测模型后,评估模型的泛化能力。原始发布未定义训练和验证分区,因此每个变体都作为单一的测试分割发布。数据集包含三个变体:1) ipknot_plus_plus:单序列结构,无多序列对齐,覆盖bpRNA-1m和Rfam 14.5目标;2) ipknot_plus_plus-ref:Rfam 14.5目标,带有Rfam参考对齐,用于共同结构预测;3) ipknot_plus_plus-mafft:Rfam 14.5目标,带有MAFFT对齐,用于共同结构预测。对齐变体覆盖相同的Rfam 14.5目标集,ID、序列和二级结构相同,仅aligned_sequences不同。数据集列包括ID、序列和二级结构(点括号表示),对齐变体额外包含对齐ID和序列列。数据集基于源数据集multimolecule/bprna和multimolecule/rfam构建,采用CC-BY-4.0许可证。
IPknot++ is a benchmark dataset for RNA secondary structure prediction, released alongside the IPknot++ paper. This dataset is intended as an external evaluation set rather than a training split, and is typically used to assess the generalization ability of RNA secondary structure prediction models after training or fine-tuning on SPOT-RNA-style splits (e.g., bpRNA-spot). The original release did not define training and validation partitions, so each variant was released as a single test split. The dataset includes three variants: 1) ipknot_plus_plus: single-sequence structure without multiple sequence alignment (MSA), covering bpRNA-1m and Rfam 14.5 targets; 2) ipknot_plus_plus-ref: Rfam 14.5 targets with Rfam reference alignments, used for consensus structure prediction; 3) ipknot_plus_plus-mafft: Rfam 14.5 targets with MAFFT alignments, used for consensus structure prediction. The alignment variants cover the same set of Rfam 14.5 targets, with identical IDs, sequences, and secondary structures, differing only in the aligned_sequences column. The dataset columns include ID, sequence, and secondary structure (dot-bracket notation), while the alignment variants additionally contain aligned ID and aligned sequence columns. This dataset is built based on the source datasets multimolecule/bprna and multimolecule/rfam, and is released under the CC-BY-4.0 license.




