KnotDiffusion Dataset

Dataset Description This dataset comprises a comprehensive collection of 207,948 monomeric knotted proteins, specifically curated for the development and evaluation of generative models for knotted proteins, such as KnotDiffusion. The collection is an aggregation of structures from two primary scientific databases: AlphaKnot 2.0: A total of 206,072 protein structures were sourced from the AlphaKnot 2.0 database. KnotProt 2.0: An additional 1,876 protein structures were obtained from the KnotProt 2.0 database. To ensure a high-quality and focused dataset for studying topological knots, a strict set of filtering criteria was applied to all selected structures: Knot Probability: A minimum knot probability of 0.8 was required. Sequence Length: The dataset is limited to proteins with sequence lengths between 60 and 512 residues. pLDDT Score (for AlphaKnot data): For structures sourced from AlphaKnot 2.0, a minimum threshold of 0.8 was applied to both the overall pLDDT (predicted Local Distance Difference Test) and the knot core pLDDT values, ensuring high-confidence model predictions. The entire dataset is provided as a single compressed archive (.tar.gz), which contains all 207,948 individual protein structures in PDB format. Users of this dataset are kindly requested to cite the original publications for the AlphaKnot 2.0 and KnotProt 2.0 databases, as detailed in the "References" section of this record, in any resulting work.