Datasetpublic

EvaKlimentova/Diffusion-all_knots

Dataset of knotted and unknotted proteins created by diffusion models This dataset supports Unveiling the Entangled Landscape of Artificial Knotted Proteins ICLR poster and related GitHub repo. The dataset contains a mixture of proteins generated by RFdiffusion+pMPNN, EvoDiff and a fraction of real proteins. For each of the three, the dataset contains 1k of knotted proteins and 4k of unknotted proteins. Each protein has: sequence label with either 0 (unknotted protein) or 1… See the full description on the dataset page: https://huggingface.co/datasets/EvaKlimentova/Diffusion-all_knots.

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes17downloads

Dataset Card

Dataset of knotted and unknotted proteins created by diffusion models

This dataset supports Unveiling the Entangled Landscape of Artificial Knotted Proteins ICLR poster and related GitHub repo.

The dataset contains a mixture of proteins generated by RFdiffusion+pMPNN, EvoDiff and a fraction of real proteins. For each of the three, the dataset contains 1k of knotted proteins and 4k of unknotted proteins.

Each protein has:

—sequence
—label with either 0 (unknotted protein) or 1 (knotted one)
—tool meaning one of the three approaches of getting the protein (RFdiffusion+pMPNN / EvoDiff / real protein)
—1024 numbers being ProtBERT-BFD (protein language model) embeddings, which were used further in our paper

Predicted structures of the artificially generated proteins can be found at https://drive.google.com/file/d/1wWqxLsFyny5ZL1BZ9asMiQ_SmpNjusUZ/view?usp=sharing