LigUnity training data

Dataset OverviewThis dataset contains protein-ligand binding affinity data organized into training and validation sets, along with supporting files for sequence alignment and temporal splitting experiments. The data is provided in both JSON and LMDB formats for efficient access.File StructureTraining Set<b>train_label.zip</b> (20.01 MB) - Contains:<code>train_label_blend_seq_full.json</code>: Primary training labels with assay information (structure as described in previous documentation)<code>train_label_pdbbind_seq.json</code>: Additional training labels from PDBBind<b>train_prot_all_blend.lmdb</b> (1.65 GB) - Protein/pocket structures for training<b>train_lig_all_blend.lmdb</b> (3.79 GB) - Ligand structures for trainingValidation Set<b>valid_label_seq.json</b> (207.89 KB) - Label file for validation set<b>valid_lig.lmdb</b> (5.96 MB) - Ligand structures for validation<b>valid_prot.lmdb</b> (5.96 MB) - Protein/pocket structures for validationSupporting Files<b>cache/</b> - Directory for training cache files (speeds up data loading)<b>align_res.zip</b> (70.44 MB) - Sequence alignment files for training HGNN<b>uniport80.clstr</b> (1.84 MB) - Protein sequence clusters (80% similarity threshold)<b>uniport40.clstr</b> (1.77 MB) - Protein sequence clusters (40% similarity threshold)<b>TIME.zip</b> - Files for split-by-time experiments (move to <code>test_datasets</code> directory in code)

数据集概述
本数据集包含蛋白-配体结合亲和力（protein-ligand binding affinity）数据，分为训练集与验证集，同时提供序列比对（sequence alignment）及时间分割实验（temporal splitting experiments）的支持文件。数据以JSON和LMDB两种格式存储，便于高效访问。

文件结构
训练集
<b>train_label.zip</b>（20.01 MB）- 包含：
<code>train_label_blend_seq_full.json</code>：含测定信息（assay information）的主要训练标签（结构详见先前文档）
<code>train_label_pdbbind_seq.json</code>：来自PDBBind的额外训练标签
<b>train_prot_all_blend.lmdb</b>（1.65 GB）- 训练用蛋白/口袋结构
<b>train_lig_all_blend.lmdb</b>（3.79 GB）- 训练用配体结构

验证集
<b>valid_label_seq.json</b>（207.89 KB）- 验证集标签文件
<b>valid_lig.lmdb</b>（5.96 MB）- 验证用配体结构
<b>valid_prot.lmdb</b>（5.96 MB）- 验证用蛋白/口袋结构

支持文件
<b>cache/</b> - 训练缓存文件目录（加速数据加载）
<b>align_res.zip</b>（70.44 MB）- 用于训练HGNN的序列比对文件
<b>uniport80.clstr</b>（1.84 MB）- 蛋白序列簇（80%相似性阈值）
<b>uniport40.clstr</b>（1.77 MB）- 蛋白序列簇（40%相似性阈值）
<b>TIME.zip</b> - 时间分割实验文件（需移至代码中的<code>test_datasets</code>目录）