Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

MitoAge: predicted structures of the 13 mitochondrial proteins in 62 mammals

806 predicted structures (62 mammals x 13 mitochondrially encoded proteins), made with ESMFold2 (biohub/ESMFold2-Fast, revision 45fe8656f5b3ef493c17fcf9abe9a2968902e712, one diffusion sample per protein, bfloat16, esm 3.4.3, torch 2.15.0) on one A100 40 GB in 11,949 s on 2026-11-15. The sequences are the ungapped rows of the MitoAge build's protein alignments; the species are those of mitoage mitoport align. They are the input of the structural-alignment check (H10 in docs/HYPOTHESES.md): FoldMason's structural alignment against MAFFT L-INS-i on the same sequences.

Part of the prepared data of dna-seq/mitoage-pipelines, the analysis code of MitoAge. The code, the hypotheses and the results are described there (README.md, docs/HYPOTHESES.md, docs/PIPELINE.md). Fetch it with uv run mitoage download prepared --topic structures in that repository: the folders land in its data/, where mitoage structure msa reads them; mitoage structure fold makes them again on a GPU.

Sister datasets: the MitoAge build, nuclear partners, embeddings, aligner variants, structures, results.

Contents

Folder What Made by
interim/structure/pdbs/<GENE>/<species-slug>.pdb one predicted structure, pLDDT in the B-factor column mitoage structure fold
interim/structure/pdbs/plddt.tsv no header: gene, species, length, mean pLDDT over CA atoms mitoage structure fold

Mean pLDDT 84.8 over the 806 models: 79 to 97 for ten proteins, 56 to 62 for ND3, ND4L and ATP8. These are predictions, not measured structures: read low-confidence regions accordingly.

Nothing here is peer reviewed; docs/HYPOTHESES.md in the code repository says which results held.

Licences

Sequences: from the MitoAge build (RefSeq, public domain; see its card). Predicted structures: CC BY 4.0. ESMFold2 is released by Biohub under the MIT licence.

Downloads last month
-

Collection including longevity-genie/mitoage-structures