multimolecule/bprna-spot-2

bpRNA-spot是一个用于RNA二级结构预测的数据集集合，由SPOT-RNA方法使用。该数据集作为复合仓库发布，包含三个编号组件：bpRNA-spot-0（初始bpRNA分割，来自bpRNA-1m的子集，通过CD-HIT去除序列相似度超过80%的序列，并随机按约8:1:1的比例分为训练、验证和测试集）、bpRNA-spot-1（用于迁移学习的高分辨率PDB RNA分割）和bpRNA-spot-2（仅包含NMR解析的RNA，用于训练后评估）。数据集整体将组件按顺序合并为训练集（TR0 + TR1）、验证集（VL0 + VL1）和测试集（TS0 + TS1 + TS2）。所有二级结构以点括号表示法存储，在转换为点括号表示法之前，移除了可能导致核苷酸与多个伙伴配对的碱基对，并将序列文件中的非A/C/G/U符号标准化为N。数据集包含id、序列、二级结构、结构注释和功能注释等列。bpRNA-spot is a collection of datasets for RNA secondary structure prediction, which is utilized by the SPOT-RNA method. This dataset is released as a composite repository, comprising three numbered components: bpRNA-spot-0, the initial bpRNA split sourced from bpRNA-1m. Sequences with >80% sequence similarity were filtered out via CD-HIT, and this subset was randomly partitioned into training, validation, and test sets at an approximate ratio of 8:1:1; bpRNA-spot-1, high-resolution PDB-derived RNA splits intended for transfer learning; and bpRNA-spot-2, which exclusively contains RNA structures solved via NMR and is designed for post-training evaluation. For the entire dataset, the components are merged sequentially to form the training set (TR0 + TR1), validation set (VL0 + VL1), and test set (TS0 + TS1 + TS2). All secondary structures are stored in dot-bracket notation. During preprocessing prior to conversion to dot-bracket notation, base pairs that would allow a single nucleotide to pair with multiple partners were removed, and non-A/C/G/U symbols in sequence files were standardized to the character N. The dataset includes columns such as id, sequence, secondary structure, structural annotation, and functional annotation.