PDB-Struct

PDB-Struct是一个全面的基于结构的蛋白质设计基准数据集，由Mila - 魁北克人工智能研究所创建。该数据集包含18,024条高质量的CATH蛋白质数据，用于评估和比较不同的蛋白质设计方法。数据集通过精心策划，包括高吞吐量的从头设计蛋白质和大规模实验突变实验数据，旨在通过引入两种新指标——基于重折叠的指标和基于稳定性的指标，来解决现有评估方法的不足。PDB-Struct不仅评估了最新的蛋白质设计模型，还对之前未比较的方法进行了评估，为未来公平和全面的蛋白质设计方法评估铺平了道路。

PDB-Struct is a comprehensive structure-based protein design benchmark dataset created by Mila – Quebec Artificial Intelligence Institute. This dataset contains 18,024 high-quality CATH protein entries, which are used to evaluate and compare different protein design methodologies. Curated meticulously, the dataset includes high-throughput de novo designed proteins and large-scale experimental mutagenesis data, aiming to address the limitations of existing evaluation approaches by introducing two novel metrics: a refolding-based metric and a stability-based metric. PDB-Struct not only evaluates state-of-the-art protein design models but also assesses previously uncompared methods, paving the way for fair and comprehensive evaluations of protein design methodologies in the future.