eczech/marinfold-exp11-protein-docs

这是一个基于AlphaFold和蛋白质语言模型的蛋白质结构预测数据集，名为“marinfold-exp11-pdocs”，是源数据集“timodonnell/protein-docs”中“contacts-and-distances-v1-5x”配置的重新发布版本。数据集按源数据中的“round”列进行质量分桶，分为high（round 0，约168万行）、medium（round 1，约142万行）和low（round 2–4，约229万行）三个配置，每个配置包含训练、验证和测试分割。数据以Parquet格式存储，保留了源数据集的所有列，包括接触图和距离信息，适用于文本生成任务，专注于英语语言，并支持蛋白质结构、AlphaFold、接触图和蛋白质语言模型等标签。数据集使用CC-BY-4.0许可证。

This is a protein structure prediction dataset based on AlphaFold and protein language models, named marinfold-exp11-pdocs, which is a re-publication of the contacts-and-distances-v1-5x config from the source dataset timodonnell/protein-docs. The dataset is quality-bucketed by the round column from the source data, divided into three configs: high (round 0, approximately 1.68M rows), medium (round 1, approximately 1.42M rows), and low (round 2–4, approximately 2.29M rows). Each config includes train, validation, and test splits. The data is stored in Parquet format, preserving all columns from the source dataset, including contact maps and distance information. It is suitable for text-generation tasks, focused on the English language, and supports tags such as protein-structure, AlphaFold, contact-map, and protein-language-model. The dataset uses the CC-BY-4.0 license.