DIPS-Plus: The Enhanced Database of Interacting Protein Structures for Interface Prediction

This dataset contains replication data for the paper titled "DIPS-Plus: The Enhanced Database of Interacting Protein Structures for Interface Prediction". The dataset consists of pickled Pandas DataFrame files, along with training, validation, and (for DB5-Plus) test filename lists for cross-validation, that can be used to develop and evaluate protein interface prediction models. This dataset also contains the externally generated residue-level PSAIA and HH-suite3 features for users' convenience (e.g. raw MSAs and profile HMMs for each protein complex). Our GitHub repository linked in the "Additional notes" metadata section below provides more details on how we parsed through these files to create our cross-validation datasets. The GitHub repository for DIPS-Plus also includes scripts that can be used to impute missing feature values and convert the final "raw" complexes into DGL-compatible graph objects. Since our final DGL graph representation for each complex uses PyTorch tensors in its construction of residue embeddings, the final representation of each complex can easily be adapted to fit the users' needs (e.g. feeding a complex's 2D residue feature tensors into a convolutional neural network).本数据集包含论文《DIPS-Plus：用于界面预测的增强型相互作用蛋白质结构数据库》的复现数据。该数据集包含序列化后的Pandas数据框（Pandas DataFrame）文件，以及用于交叉验证的训练集、验证集和（针对DB5-Plus的）测试集文件名列表，可用于开发与评估蛋白质界面预测模型。为方便用户使用，本数据集还包含外部生成的残基水平PSAIA与HH-suite3特征（例如每个蛋白质复合物的原始多序列比对结果（MSAs）及隐马尔可夫模型剖面（profile HMMs））。下文"附加说明"元数据部分中链接的GitHub仓库，详细说明了我们如何解析这些文件以构建交叉验证数据集。DIPS-Plus的GitHub仓库还附带了相关脚本，可用于补全缺失的特征值，并将最终的"原始"复合物转换为适配DGL的图对象。由于我们针对每个复合物的最终DGL图表示在构建残基嵌入时采用了PyTorch张量，因此每个复合物的最终表示可轻松适配用户的各类需求（例如将复合物的二维残基特征张量输入卷积神经网络）。