Datasets with embeddings and other representations for all proteins in Uniprot/Swiss-Prot
See the codeScientific data lake with PLM embeddings, GO annotations and taxonomy representations for all proteins in Uniprot/Swiss-Prot
Proteins are sorted by length. All files contain the same sequence of proteins, so the "ids.txt" file can be used as the row names.
Several models are used to create computational descriptions of the Swiss-Prot proteins:
| Name | Model 🤖 | Vector Length 📏 | File Size | Download Links 🔗 |
|---|---|---|---|---|
| emb.prottrans.parquet | prottrans_t5_xl_u50 (calculated by Uniprot) | 1024 | 1.3G | HF, UFRN |
| emb.ankh_large.parquet | ankh-large | 1536 | 3.4G | HF, UFRN |
| emb.ankh_base.parquet | ankh-base | 768 | 1.7G | HF, UFRN |
| emb.esm2_t36.parquet | esm2_t36_3B_UR50D | 2560 | 5.7G | HF, UFRN |
| emb.esm2_t33.parquet | esm2_t33_650M_UR50D | 1280 | 2.8G | HF, UFRN |
| emb.esm2_t30.parquet | esm2_t30_150M_UR50D | 640 | 1.4G | HF, UFRN |
| emb.esm2_t12.parquet | esm2_t12_35M_UR50D | 480 | 1G | HF, UFRN |
| emb.esm2_t6.parquet | esm2_t6_8M_UR50D | 320 | 700M | HF, UFRN |
| Name | Content | Download Links 🔗 |
|---|---|---|
| ids.txt | Uniprot Accession IDs | UFRN |
| uniprot_sorted.fasta.gz | Aminoacid sequences of SwissProt proteins | UFRN |
| taxid.tsv | NCBI taxon ID of each protein | UFRN |
All Gene Ontology annotations of Swiss-Prot proteins, excluding computational, non-traceable and no-data annotations. The full list of ignored evidence codes is available at evi_not_to_use.txt. Annotations have been "expanded upwards": parent terms of existing annotations have been included in these files.
| Name | Content | Download Links 🔗 |
|---|---|---|
| go.expanded.tsv.gz | MF, BP and CC annotations in simplified GAF format | UFRN |
| go.experimental.mf.tsv.gz | Molecular Functions | UFRN |
| go.experimental.bp.tsv.gz | Biological Processes | UFRN |
| go.experimental.cc.tsv.gz | Cellular Components | UFRN |
Numerical representations of the NCBI taxon IDs of each protein. Instead of the original NCBI taxonomy tree, we use the custom taxonomy created by taxallnomy project, because it attributes the same number of parent taxa (genus, family, order...) to each species ID.
| Name | Description | Vector Length 📏 | Download Links 🔗 |
|---|---|---|---|
| emb.taxa_profile_256.parquet | Taxa Proximity [0.0, 1.0] to each one of the 256 most annotated taxa | 256 | UFRN |
| emb.taxa_profile_128.parquet | Taxa Proximity [0.0, 1.0] to each one of the 128 most annotated taxa | 128 | UFRN |
| onehot.taxa_256.parquet | Taxa One-Hot Encoding | 256 | HF, UFRN |
| onehot.taxa_128.parquet | Taxa One-Hot Encoding | 128 | HF, UFRN |
| Files | Format Descriptions |
|---|---|
| ids.txt | One UniprotID per line |
| taxid.tsv | Tab-separated table with columns: UniprotID, NCBI Taxon ID |
| go.expanded.tsv.gz | Tab-separated table with columns: UniprotID, GO ID, ECO ID, NCBI Taxon ID, GO Ontology Code |
| go.experimental.*.tsv.gz | Tab-separated table with columns: UniprotID, GO IDs separated by ',' |
| *.parquet | Parquet formatted dataset. Has only two columns ('id' and 'emb'). For rows where an embedding could not be defined, a vector of np.NaN is placed. |
Requirements to generate the datasets from scratch:
Test:
$ mkdir test
$ nextflow run main.nf --mode test --release_dir test
Full release:
$ mkdir <path to generate database at>
$ nextflow run main.nf --mode release --release_dir <path to generate database at>
Please cite the following work:
Bibtext:
@inproceedings{AlvesSobrinho2025ProteinDimensionDB,
author = {Pit{\'{a}}goras de Azevedo Alves Sobrinho and Tetsu Sakamoto and Wilfredo Blanco Figuerola},
title = {Protein Dimension DB: A Unified Protein Repository for Representation Learning and Functional Analysis},
booktitle = {BioInformatics: 21st Brazilian Congress, X-Meeting 2025, João Pessoa, Brazil, June 3–6, 2025, Proceedings},
series = {Lecture Notes in Computer Science},
volume = {16037},
year = {2025},
editor = {Marcio Dorn and Fabricio Martins Lopes},
publisher = {Springer Cham},
isbn = {978-3-032-09335-6},
eisbn = {978-3-032-09336-3},
address = {Cham, Switzerland}
}
APA reference:
Alves Sobrinho, P. de A., Sakamoto, T., & Blanco Figuerola, W. (2025). Protein Dimension DB: A unified protein repository for representation learning and functional analysis. BioInformatics: 21st Brazilian Congress, X-Meeting 2025, João Pessoa, Brazil, June 3–6, 2025, Proceedings (Lecture Notes in Computer Science, Vol. 16037). Springer Cham.
Datasets with embeddings and other representations for all proteins in Uniprot/Swiss-Prot
See the codeScientific data lake with PLM embeddings, GO annotations and taxonomy representations for all proteins in Uniprot/Swiss-Prot
Proteins are sorted by length. All files contain the same sequence of proteins, so the "ids.txt" file can be used as the row names.
Several models are used to create computational descriptions of the Swiss-Prot proteins:
| Name | Model 🤖 | Vector Length 📏 | File Size | Download Links 🔗 |
|---|---|---|---|---|
| emb.prottrans.parquet | prottrans_t5_xl_u50 (calculated by Uniprot) | 1024 | 1.3G | HF, UFRN |
| emb.ankh_large.parquet | ankh-large | 1536 | 3.4G | HF, UFRN |
| emb.ankh_base.parquet | ankh-base | 768 | 1.7G | HF, UFRN |
| emb.esm2_t36.parquet | esm2_t36_3B_UR50D | 2560 | 5.7G | HF, UFRN |
| emb.esm2_t33.parquet | esm2_t33_650M_UR50D | 1280 | 2.8G | HF, UFRN |
| emb.esm2_t30.parquet | esm2_t30_150M_UR50D | 640 | 1.4G | HF, UFRN |
| emb.esm2_t12.parquet | esm2_t12_35M_UR50D | 480 | 1G | HF, UFRN |
| emb.esm2_t6.parquet | esm2_t6_8M_UR50D | 320 | 700M | HF, UFRN |
| Name | Content | Download Links 🔗 |
|---|---|---|
| ids.txt | Uniprot Accession IDs | UFRN |
| uniprot_sorted.fasta.gz | Aminoacid sequences of SwissProt proteins | UFRN |
| taxid.tsv | NCBI taxon ID of each protein | UFRN |
All Gene Ontology annotations of Swiss-Prot proteins, excluding computational, non-traceable and no-data annotations. The full list of ignored evidence codes is available at evi_not_to_use.txt. Annotations have been "expanded upwards": parent terms of existing annotations have been included in these files.
| Name | Content | Download Links 🔗 |
|---|---|---|
| go.expanded.tsv.gz | MF, BP and CC annotations in simplified GAF format | UFRN |
| go.experimental.mf.tsv.gz | Molecular Functions | UFRN |
| go.experimental.bp.tsv.gz | Biological Processes | UFRN |
| go.experimental.cc.tsv.gz | Cellular Components | UFRN |
Numerical representations of the NCBI taxon IDs of each protein. Instead of the original NCBI taxonomy tree, we use the custom taxonomy created by taxallnomy project, because it attributes the same number of parent taxa (genus, family, order...) to each species ID.
| Name | Description | Vector Length 📏 | Download Links 🔗 |
|---|---|---|---|
| emb.taxa_profile_256.parquet | Taxa Proximity [0.0, 1.0] to each one of the 256 most annotated taxa | 256 | UFRN |
| emb.taxa_profile_128.parquet | Taxa Proximity [0.0, 1.0] to each one of the 128 most annotated taxa | 128 | UFRN |
| onehot.taxa_256.parquet | Taxa One-Hot Encoding | 256 | HF, UFRN |
| onehot.taxa_128.parquet | Taxa One-Hot Encoding | 128 | HF, UFRN |
| Files | Format Descriptions |
|---|---|
| ids.txt | One UniprotID per line |
| taxid.tsv | Tab-separated table with columns: UniprotID, NCBI Taxon ID |
| go.expanded.tsv.gz | Tab-separated table with columns: UniprotID, GO ID, ECO ID, NCBI Taxon ID, GO Ontology Code |
| go.experimental.*.tsv.gz | Tab-separated table with columns: UniprotID, GO IDs separated by ',' |
| *.parquet | Parquet formatted dataset. Has only two columns ('id' and 'emb'). For rows where an embedding could not be defined, a vector of np.NaN is placed. |
Requirements to generate the datasets from scratch:
Test:
$ mkdir test
$ nextflow run main.nf --mode test --release_dir test
Full release:
$ mkdir <path to generate database at>
$ nextflow run main.nf --mode release --release_dir <path to generate database at>
Please cite the following work:
Bibtext:
@inproceedings{AlvesSobrinho2025ProteinDimensionDB,
author = {Pit{\'{a}}goras de Azevedo Alves Sobrinho and Tetsu Sakamoto and Wilfredo Blanco Figuerola},
title = {Protein Dimension DB: A Unified Protein Repository for Representation Learning and Functional Analysis},
booktitle = {BioInformatics: 21st Brazilian Congress, X-Meeting 2025, João Pessoa, Brazil, June 3–6, 2025, Proceedings},
series = {Lecture Notes in Computer Science},
volume = {16037},
year = {2025},
editor = {Marcio Dorn and Fabricio Martins Lopes},
publisher = {Springer Cham},
isbn = {978-3-032-09335-6},
eisbn = {978-3-032-09336-3},
address = {Cham, Switzerland}
}
APA reference:
Alves Sobrinho, P. de A., Sakamoto, T., & Blanco Figuerola, W. (2025). Protein Dimension DB: A unified protein repository for representation learning and functional analysis. BioInformatics: 21st Brazilian Congress, X-Meeting 2025, João Pessoa, Brazil, June 3–6, 2025, Proceedings (Lecture Notes in Computer Science, Vol. 16037). Springer Cham.