training data for SoleNNoID protein annotation software

This dataset represents an edited and upgraded version of the REPETITA dataset which was used to train and test the SOLeNNoID convolutional neural network for solenoid residue classification. The original dataset comprised PDB files with full or partial protein structures for solenoid and non-solenoid proteins, as well as residue spans (e.g. residue 224-280) for solenoid regions. In this version of the dataset, we include the original PDB files, distance matrices derived from these PDB files, which account for the missing residues in the PDB files, as well as revised labels by mapping and editing the original residue spans to produce a per-residue label file to match each structure/distance matrix. Additionally, extra beta-solenoid entries were manually added to the dataset. Finally, a new test dataset was curated from reviewed solenoid entries in the RepeatsDB database. The dataset is split into training_validation_dataset and test_dataset directories. The training_validation_dataset directory comprises the original REPETITA dataset, split into alpha-, alpha/beta-, beta-, and non-solenoid directories. Each of these directories contains subdirectories with distance matrices, labels, and PDB structures. In addition, the beta_additional directory contains subdirectories with the distance matrices, labels and PDB structures of further manually added beta-solenoid entries. The test_dataset directory comprises directories with non-solenoid and solenoid structures in mmCIF format, as well as a directory with the ground truth labels, and labels predicted by the TAPO, RepeatsDB-Lite and PRIGSA2 methods. References and links: REPETITA: Luca Marsella, Francesco Sirocco, Antonio Trovato, Flavio Seno, Silvio C.E. Tosatto, REPETITA: detection and discrimination of the periodicity of protein solenoid repeats by discrete Fourier transform, Bioinformatics, Volume 25, Issue 12, June 2009, Pages i289–i295, https://doi.org/10.1093/bioinformatics/btp232 RAPHAEL: Ian Walsh, Francesco G. Sirocco, Giovanni Minervini, Tomás Di Domenico, Carlo Ferrari, Silvio C. E. Tosatto, RAPHAEL: recognition, periodicity and insertion assignment of solenoid protein structures, Bioinformatics, Volume 28, Issue 24, December 2012, Pages 3257–3264, https://doi.org/10.1093/bioinformatics/bts550 SOLeNNoID: Nikov, Georgi and Pretorius, Daniella and Murray, James W., SOLeNNoID: A Deep Learning Pipeline For Solenoid Residue Detection in Protein Structures, bioRxiv, 2024 REPETITA/RAPHAEL dataset link: http://old.protein.bio.unipd.it/raphael/precompiled.html本数据集为REPETITA数据集的编辑升级版本，该数据集曾用于训练与测试用于螺线管残基分类的SOLeNNoID卷积神经网络（convolutional neural network）。原始数据集包含针对螺线管蛋白与非螺线管蛋白的完整或部分蛋白质结构的PDB（Protein Data Bank）文件，同时涵盖螺线管区域的残基跨度信息（例如残基224-280）。 在本数据集版本中，我们保留了原始PDB文件，以及从这些PDB文件中推导得到的距离矩阵——该矩阵已对PDB文件中的缺失残基进行了补偿；同时还包含经修订的标签：通过映射与编辑原始残基跨度信息，生成了与每个结构/距离矩阵匹配的逐残基标签文件。此外，我们还手动向数据集中添加了额外的β-螺线管条目。 最后，本数据集从RepeatsDB数据库中经过审核的螺线管条目中整理得到了全新的测试集。 本数据集分为training_validation_dataset与test_dataset两个目录。 training_validation_dataset目录包含原始REPETITA数据集，并将其进一步划分为alpha-、alpha/beta-、beta-以及non-solenoid四个子目录。每个子目录均包含存放距离矩阵、标签与PDB结构的子文件夹。此外，beta_additional目录中包含存放额外手动添加的β-螺线管条目的距离矩阵、标签与PDB结构的子文件夹。 test_dataset目录包含存放mmCIF格式非螺线管与螺线管结构的目录，以及存放基准真实标签，以及TAPO、RepeatsDB-Lite与PRIGSA2三种方法预测得到的标签的目录。 参考文献与链接： REPETITA：Luca Marsella、Francesco Sirocco、Antonio Trovato、Flavio Seno、Silvio C.E. Tosatto，REPETITA：基于离散傅里叶变换检测与区分蛋白质螺线管重复序列的周期性，《生物信息学》，第25卷第12期，2009年6月，页码i289–i295，https://doi.org/10.1093/bioinformatics/btp232 RAPHAEL：Ian Walsh、Francesco G. Sirocco、Giovanni Minervini、Tomás Di Domenico、Carlo Ferrari、Silvio C. E. Tosatto，RAPHAEL：螺线管蛋白质结构的识别、周期性分析与插入位点分配，《生物信息学》，第28卷第24期，2012年12月，页码3257–3264，https://doi.org/10.1093/bioinformatics/bts550 SOLeNNoID：Nikov Georgi、Pretorius Daniella、Murray James W.，SOLeNNoID：用于蛋白质结构中螺线管残基检测的深度学习流程，bioRxiv，2024年 REPETITA/RAPHAEL数据集链接：http://old.protein.bio.unipd.it/raphael/precompiled.html