EvaKlimentova/knots_AF

Dataset of knotted and unknotted proteins from AlphaFold

This dataset was created based on the whole UniProt database from where protein sequences were obtained, their 3D structure computed with AlphaFold and their topology (knot vs. unknot) determined with Topoly. The obtained proteins were further processed by clustering with CD-HIT and downsampling to ensure balanced sequence length distribution between knotted and unknotted samples.

The final dataset contains 98528 knotted proteins and the same amount of unknotted ones. It's split into train set with 157644 (80%) samples and test datasets with 39412 (20%) samples.