Hey RNA folks!
\nIf you've been training models with the officially released datasets but want to challenge your models in a more out of distribution setting, I've released all RNA 3D structures from the PDB clustered by both sequence and structural similarity. This is part of the gRNAde project for 3D RNA structure design.
\nThe data was prepared using RNASolo, a repository of all known RNA 3D structures from the PDB extracted from solo RNAs as well as biomolecular complexes.
\nProcessed data in machine learning-ready format (slightly different from this contest) is available here: https://drive.google.com/file/d/1gcUUaRxbGZnGMkLdtVwAILWVerVCbu4Y/view
\nThe attached jupyter notebook shows you how to work with the data -- hopefully it should be very easy to construct a csv file in the same format as this contest. The notebook links to download processed gRNAde data, access cluster IDs, and handing missing nucleotides and/or missing coordinates.
\nAnother good source of external data is RNA3DB, which provides the same functionality with even more rigorous pre-processing and preparation.
", "messages": [ { "id": 3194342, "postDate": "2025-05-05T18:08:19.427Z", "content": "Hey RNA folks!
\nIf you've been training models with the officially released datasets but want to challenge your models in a more out of distribution setting, I've released all RNA 3D structures from the PDB clustered by both sequence and structural similarity. This is part of the gRNAde project for 3D RNA structure design.
\nThe data was prepared using RNASolo, a repository of all known RNA 3D structures from the PDB extracted from solo RNAs as well as biomolecular complexes.
\nProcessed data in machine learning-ready format (slightly different from this contest) is available here: https://drive.google.com/file/d/1gcUUaRxbGZnGMkLdtVwAILWVerVCbu4Y/view
\nThe attached jupyter notebook shows you how to work with the data -- hopefully it should be very easy to construct a csv file in the same format as this contest. The notebook links to download processed gRNAde data, access cluster IDs, and handing missing nucleotides and/or missing coordinates.
\nAnother good source of external data is RNA3DB, which provides the same functionality with even more rigorous pre-processing and preparation.
", "rawMarkdown": "Hey RNA folks!\n\nIf you've been training models with the officially released datasets but want to challenge your models in a more out of distribution setting, I've released **all RNA 3D structures from the PDB** clustered by both sequence and structural similarity. This is part of the [gRNAde](https://github.com/chaitjo/geometric-rna-design) project for 3D RNA structure design.\n\nThe data was prepared using [RNASolo](https://rnasolo.cs.put.poznan.pl/), a repository of all known RNA 3D structures from the PDB extracted from solo RNAs as well as biomolecular complexes.\n\nProcessed data in machine learning-ready format (slightly different from this contest) is available here: https://drive.google.com/file/d/1gcUUaRxbGZnGMkLdtVwAILWVerVCbu4Y/view \n\nThe attached jupyter notebook shows you how to work with the data -- hopefully it **should be very easy to construct a csv file** in the same format as this contest. The notebook links to download processed gRNAde data, access cluster IDs, and handing missing nucleotides and/or missing coordinates.\n- There are **over 12 thousand** 3D structures from 4.2K unique RNA sequences.\n- The processed coordinates use an Atom27 format where each nucleotide has a subset of 27 possible atoms. The atoms are ordered in a pre-defined manner which is specified in the notebook.\n- There are 4223 unique sequences, and 1102 structural clusters determined by USalign with a cutoff of 0.45 TM score.\n- And 1002 clusters grouped by sequence identity at a cutoff 0.8 determined by CD-Hit.\n\nAnother good source of external data is [RNA3DB](https://github.com/marcellszi/rna3db), which provides the same functionality with even more rigorous pre-processing and preparation.", "votes": 10 } ], "comments": [], "raw_markdown_by_id": { "3194342": "Hey RNA folks!\n\nIf you've been training models with the officially released datasets but want to challenge your models in a more out of distribution setting, I've released **all RNA 3D structures from the PDB** clustered by both sequence and structural similarity. This is part of the [gRNAde](https://github.com/chaitjo/geometric-rna-design) project for 3D RNA structure design.\n\nThe data was prepared using [RNASolo](https://rnasolo.cs.put.poznan.pl/), a repository of all known RNA 3D structures from the PDB extracted from solo RNAs as well as biomolecular complexes.\n\nProcessed data in machine learning-ready format (slightly different from this contest) is available here: https://drive.google.com/file/d/1gcUUaRxbGZnGMkLdtVwAILWVerVCbu4Y/view \n\nThe attached jupyter notebook shows you how to work with the data -- hopefully it **should be very easy to construct a csv file** in the same format as this contest. The notebook links to download processed gRNAde data, access cluster IDs, and handing missing nucleotides and/or missing coordinates.\n- There are **over 12 thousand** 3D structures from 4.2K unique RNA sequences.\n- The processed coordinates use an Atom27 format where each nucleotide has a subset of 27 possible atoms. The atoms are ordered in a pre-defined manner which is specified in the notebook.\n- There are 4223 unique sequences, and 1102 structural clusters determined by USalign with a cutoff of 0.45 TM score.\n- And 1002 clusters grouped by sequence identity at a cutoff 0.8 determined by CD-Hit.\n\nAnother good source of external data is [RNA3DB](https://github.com/marcellszi/rna3db), which provides the same functionality with even more rigorous pre-processing and preparation." } }