PaperBLAST

Or see example results for the putative alcohol dehydrogenase WP_012018426.1, which is actually the regulator ercA.

Related Tools

• SitesBLAST and Sites on a Tree
• Curated BLAST for Genomes
• GapMind for amino acid biosynthesis or carbon catabolism
• Family Search vs. Papers
• Papers vs. PDB
• Papers vs. PFam
  • DUFs with characterized representatives
  • Characterized proteins not in PFam

Statistics

How It Works

PaperBLAST builds a database of protein sequences that are linked to scientific articles. These links come from automated text searches against the articles in EuropePMC and from manually-curated information from GeneRIF, UniProtKB/Swiss-Prot, BRENDA, CAZy (as made available by dbCAN), BioLiP, CharProtDB, MetaCyc, EcoCyc, TCDB, REBASE, the Fitness Browser, and a subset of the European Nucleotide Archive with the /experiment tag. Given this database and a protein sequence query, PaperBLAST uses protein-protein BLAST to find similar sequences with E < 0.001.

To build the database, we query EuropePMC with locus tags, with RefSeq protein identifiers, and with UniProt accessions. We obtain the locus tags from RefSeq or from MicrobesOnline. We use queries of the form "locus_tag AND genus_name" to try to ensure that the paper is actually discussing that gene. Because EuropePMC indexes most recent biomedical papers, even if they are not open access, some of the links may be to papers that you cannot read or that our computers cannot read. We query each of these identifiers that appears in the open access part of EuropePMC, as well as every locus tag that appears in the 500 most-referenced genomes, so that a gene may appear in the PaperBLAST results even though none of the papers that mention it are open access. We also incorporate text-mined links from EuropePMC that link open access articles to UniProt or RefSeq identifiers. (This yields some additional links because EuropePMC uses different heuristics for their text mining than we do.)

For every article that mentions a locus tag, a RefSeq protein identifier, or a UniProt accession, we try to select one or two snippets of text that refer to the protein. If we cannot get access to the full text, we try to select a snippet from the abstract, but unfortunately, unique identifiers such as locus tags are rarely provided in abstracts.

PaperBLAST also incorporates manually-curated protein functions:

• Proteins from NCBI's RefSeq are included if a GeneRIF entry links the gene to an article in PubMed^®. GeneRIF also provides a short summary of the article's claim about the protein, which is shown instead of a snippet.
• Proteins from Swiss-Prot (the curated part of UniProt) are included if the curators identified experimental evidence for the protein's function (evidence code ECO:0000269). For these proteins, the fields of the Swiss-Prot entry that describe the protein's function are shown (with bold headings).
• Proteins from BRENDA, a curated database of enzymes, are included if they are linked to a paper in PubMed and their full sequence is known.
• Every protein from the non-redundant subset of BioLiP, a database of ligand-binding sites and catalytic residues in protein structures, is included. Since BioLiP itself does not include descriptions of the proteins, those are taken from the Protein Data Bank. Descriptions from PDB rely on the original submitter of the structure and cannot be updated by others, so they may be less reliable. (For SitesBLAST and Sites on a Tree, we use a larger subset of BioLiP so that every ligand is represented among a group of structures with similar sequences, but for PaperBLAST, we use the non-redundant set provided by BioLiP.)
• Every protein from EcoCyc, a curated database of the proteins in Escherichia coli K-12, is included, regardless of whether they are characterized or not.
• Proteins from the MetaCyc metabolic pathway database are included if they are linked to a paper in PubMed and their full sequence is known.
• Proteins from the Transport Classification Database (TCDB) are included if they have known substrate(s), have reference(s), and are not described as uncharacterized or putative. (Some of the references are not visible on the PaperBLAST web site.)
• Every protein from CharProtDB, a database of experimentally characterized protein annotations, is included.
• Proteins from the CAZy database of carbohydrate-active enzymes are included if they are associated with an Enzyme Classification number. Even though CAZy does not provide links from individual protein sequences to papers, these should all be experimentally-characterized proteins.
• Proteins from the REBASE database of restriction enzymes are included if they have known specificity.
• Every protein with an evidence-based reannotation (based on mutant phenotypes) in the Fitness Browser is included.
• Sequence-specific transcription factors (including sigma factors and DNA-binding response regulators) with experimentally-determined DNA binding sites from the PRODORIC database of gene regulation in prokaryotes.
• Putative transcription factors from RegPrecise that have manually-curated predictions for their binding sites. These predictions are based on conserved putative regulatory sites across genomes that contain similar transcription factors, so PaperBLAST clusters the TFs at 70% identity and retains just one member of each cluster.
• Coding sequence (CDS) features from the European Nucleotide Archive (ENA) are included if the /experiment tag is set (implying that there is experimental evidence for the annotation), the nucleotide entry links to paper(s) in PubMed, and the nucleotide entry is from the STD data class (implying that these are targeted annotated sequences, not from shotgun sequencing). Also, to filter out genes whose transcription or translation was detected, but whose function was not studied, nucleotide entries or papers with more than 25 such proteins are excluded. Descriptions from ENA rely on the original submitter of the sequence and cannot be updated by others, so they may be less reliable.

Except for GeneRIF and ENA, the curated entries include a short curated description of the protein's function. For entries from BioLiP, the protein's function may not be known beyond binding to the ligand. Many of these entries also link to articles in PubMed.

For more information see the PaperBLAST paper (mSystems 2017) or the code. You can download PaperBLAST's database here.

Changes to PaperBLAST since the paper was written:

• November 2023: incorporated PRODORIC and RegPrecise. Many PRODORIC entries were not linked to a protein sequence (no UniProt identifier), so we added this information.
• February 2023: BioLiP changed their download format. PaperBLAST now includes their non-redundant subset. SitesBLAST and Sites on a Tree use a larger non-redundant subset that ensures that every ligand is represented within each cluster. This should ensure that every binding site is represented.
• June 2022: incorporated some coding sequences from ENA with the /experiment tag.
• March 2022: incorporated BioLiP.
• April 2020: incorporated TCDB.
• April 2019: EuropePMC now returns table entries in their search results. This has expanded PaperBLAST's database, but most of the new entries are of low relevance, and the resulting snippets are often just lists of locus tags with annotations.
• February 2018: the alignment page reports the conservation of the hit's functional sites (if available from from Swiss-Prot or UniProt)
• January 2018: incorporated BRENDA.
• December 2017: incorporated MetaCyc, CharProtDB, CAZy, REBASE, and the reannotations from the Fitness Browser.
• September 2017: EuropePMC no longer returns some table entries in their search results. This has shrunk PaperBLAST's database, but has also reduced the number of low-relevance hits.

Many of these changes are described in Interactive tools for functional annotation of bacterial genomes.

Secrets

PaperBLAST cannot provide snippets for many of the papers that are published in non-open-access journals. This limitation applies even if the paper is marked as "free" on the publisher's web site and is available in PubmedCentral or EuropePMC. If a journal that you publish in is marked as "secret," please consider publishing elsewhere.

Omissions from the PaperBLAST Database

Many important articles are missing from PaperBLAST, either because the article's full text is not in EuropePMC (as for many older articles), or because the paper does not mention a protein identifier such as a locus tag, or because of PaperBLAST's heuristics. If you notice an article that characterizes a protein's function but is missing from PaperBLAST, please notify the curators at UniProt or add an entry to GeneRIF. Entries in either of these databases will eventually be incorporated into PaperBLAST. Note that to add an entry to UniProt, you will need to find the UniProt identifier for the protein. If the protein is not already in UniProt, you can ask them to create an entry. To add an entry to GeneRIF, you will need an NCBI Gene identifier, but unfortunately many prokaryotic proteins in RefSeq do not have corresponding Gene identifers.

References

PaperBLAST: Text-mining papers for information about homologs.
M. N. Price and A. P. Arkin (2017). mSystems, 10.1128/mSystems.00039-17.

Europe PMC in 2017.
M. Levchenko et al (2017). Nucleic Acids Research, 10.1093/nar/gkx1005.

Gene indexing: characterization and analysis of NLM's GeneRIFs.
J. A. Mitchell et al (2003). AMIA Annu Symp Proc 2003:460-464.

UniProt: the universal protein knowledgebase.
The UniProt Consortium (2016). Nucleic Acids Research, 10.1093/nar/gkw1099.

BRENDA in 2017: new perspectives and new tools in BRENDA.
S. Placzek et al (2017). Nucleic Acids Research, 10.1093/nar/gkw952.

The EcoCyc database: reflecting new knowledge about Escherichia coli K-12.
I. M. Keeseler et al (2016). Nucleic Acids Research, 10.1093/nar/gkw1003.

The MetaCyc database of metabolic pathways and enzymes.
R. Caspi et al (2018). Nucleic Acids Research, 10.1093/nar/gkx935.

CharProtDB: a database of experimentally characterized protein annotations.
R. Madupu et al (2012). Nucleic Acids Research, 10.1093/nar/gkr1133.

The carbohydrate-active enzymes database (CAZy) in 2013.
V. Lombard et al (2014). Nucleic Acids Research, 10.1093/nar/gkt1178.

The Transporter Classification Database (TCDB): recent advances
M. H. Saier, Jr. et al (2016). Nucleic Acids Research, 10.1093/nar/gkv1103.

REBASE - a database for DNA restriction and modification: enzymes, genes and genomes.
R. J. Roberts et al (2015). Nucleic Acids Research, 10.1093/nar/gku1046.

Deep annotation of protein function across diverse bacteria from mutant phenotypes.
M. N. Price et al (2016). bioRxiv, 10.1101/072470.