multimolecule/bprna-spot-0

bpRNA-spot是一个用于RNA二级结构预测的数据集集合，由SPOT-RNA使用。该数据集作为一个复合存储库发布，包含三个编号的组件存储库：bpRNA-spot-0（初始bpRNA分割，包括TR0、VL0和TS0）、bpRNA-spot-1（PDB迁移学习分割，包括TR1、VL1和TS1）和bpRNA-spot-2（仅NMR评估分割，包括TS2）。这些组件按顺序拼接成训练集（TR0 + TR1）、验证集（VL0 + VL1）和测试集（TS0 + TS1 + TS2）。TR0/VL0/TS0分割是bpRNA-1m的子集，通过CD-HIT（CD-HIT-EST）去除序列相似性超过80%的序列，并将剩余序列按约8:1:1的比例随机分割为训练、验证和测试集。TR1/VL1/TS1分割包含用于迁移学习的高分辨率PDB RNA。TS2分割包含39个通过NMR解析的RNA，用于后训练评估。所有二级结构均以点括号符号存储。在序列/标签分割中，在转换为点括号符号之前，会移除可能导致核苷酸与多个伙伴配对的碱基对。序列文件中的非A/C/G/U符号被归一化为N。数据集包含以下列：id（序列标识符）、sequence（RNA序列）、secondary_structure（点括号符号表示的二级结构，可能使用超出()的括号层级表示假结）、structural_annotation（基于存储的点括号结构生成的bpRNA样式结构注释）和functional_annotation（基于存储的点括号结构生成的bpRNA样式功能注释）。该数据集主要用于文本生成和掩码语言建模任务，支持语言建模和掩码语言建模。bpRNA-spot is a collection of datasets for RNA secondary structure prediction used by SPOT-RNA. Released as a composite repository, it contains three numbered component repositories: bpRNA-spot-0 (the initial bpRNA split including TR0, VL0 and TS0), bpRNA-spot-1 (the PDB transfer learning split including TR1, VL1 and TS1) and bpRNA-spot-2 (the NMR-only evaluation split including TS2). These components are concatenated to form the training set (TR0 + TR1), validation set (VL0 + VL1) and test set (TS0 + TS1 + TS2). The TR0/VL0/TS0 split is a subset of bpRNA-1m, where sequences with over 80% sequence similarity are removed using CD-HIT (CD-HIT-EST), and the remaining sequences are randomly split into training, validation and test sets at an approximate ratio of 8:1:1. The TR1/VL1/TS1 split contains high-resolution PDB RNAs for transfer learning. The TS2 split includes 39 NMR-resolved RNAs for post-training evaluation. All secondary structures are stored in dot-bracket notation. During sequence/label splitting, base pairs that could cause a nucleotide to pair with multiple partners are removed prior to conversion to dot-bracket notation. Non-A/C/G/U symbols in sequence files are normalized to N. The dataset includes the following columns: id (sequence identifier), sequence (RNA sequence), secondary_structure (secondary structure represented by dot-bracket notation, where bracket layers beyond standard parentheses may be used to denote pseudoknots), structural_annotation (bpRNA-style structural annotations generated from the stored dot-bracket structure) and functional_annotation (bpRNA-style functional annotations generated from the stored dot-bracket structure). This dataset is primarily used for text generation and masked language modeling tasks, supporting both language modeling and masked language modeling.