A new InterPro member database

CDD joins InterPro

We are pleased to announce that the NCBI Conserved Domain Database (CDD) has joined the InterPro consortium as a member database, and has begun to be integrated into the resource.  This is the first new member database to be integrated into InterPro since HAMAP was included, back in 2009.  As you can see, it has been 7 years since a new database has been added, and this is a first for the current InterPro team.

While there are some similarities between CDD and InterPro, with both resource aggregating data from Pfam, TIGRFAMs and SMART, CDD also utilises COGs and PRKs, whereas InterPro incorporates eight other resources.  Unlike InterPro, the CDD team also curate their own models, using position specific scoring matrixes (PSSMs) to represent protein domains, and it is these models that have been prioritised for integration into InterPro.

What does CDD bring?

This release contains all 11,273 CDD models, with 318 already integrated into InterPro.  It will take time for the rest of the entries to be curated into the InterPro hierarchy, ensuring that we are consistent with both CDD’s and our own relationship trees, as well as assigning GO terms for InterPro2GO. The NCBI models are often functionally specific, with multiple CDD entries covering the same sequence set as a single Pfam profile hidden Markov model (HMM), for example.

Having such algorithmic and database diversity helps capture as much knowledge as possible about the function of a protein.  CDD uses a derivative of RPS-BLAST that performs the appropriate assignment of proteins to database entries. This software, rpsbproc (also known as CD-search), has been incorporated into the latest version of InterProScan, making CDD accessible via our web services too.

Figure 1. InterProScan result for UniProt protein A0AM81. CDD entry cd00770 extends coverage of the protein and adds more functionally specific information about the C-terminal aminoacyl-tRNA synthetase domain (in this case, that it is a serine-tRNA ligase domain).

What Next?

With summer fast approaching, we have our typical break from database releases, with the next InterPro release not anticipated until the beginning of September.  In the meantime, we have a number of planned infrastructural changes that will broaden the scope of InterPro further... We will describe these changes in detail once they are ready!

Rob Finn
on behalf of the InterPro team

Navigating the ever-changing ocean of biological knowledge

The removal of annotation from biological databases is often taken to mean that the annotation was wrong in the first place. Why else would diligent biocuators remove information that had been painstakingly added to database entries? In our recent paper, 'GO annotation in InterPro: why stability does not indicate accuracy in a sea of changing annotations', we look at some of the diverse, data-driven changes that can underlie the deletion or update of Gene Ontology annotations in the InterPro database, and highlight some of the consequent effects of these changes on UniProt protein annotations. We also explain why these changes don't necessarily mean that the original annotations were unreliable. Alternatively, we argue that they signify a curation effort committed to annotation accuracy, attempting to navigate an ever-changing ocean of biological knowledge.

Alex Mitchell
on behalf of the InterPro team

What's new in InterPro release 50.0 and 51.0

Faster InterPro member database processing:

InterPro releases 50.0 and 51.0 have brought some important developments from an InterPro production point of view, which we thought would be worth sharing. Release 50.0 saw the incorporation of a new version of PIRSF, which has importantly been migrated to use the HMMER3.1b analysis algorithm. This version of HMMER runs approximately one thousand times faster than the previous version used by PIRSF (HMMER2.0), helping to ensure that InterPro can continue to calculate UniProtKB match data in a timely manner. In a related development, as part of InterPro release 51.0, we debuted a sequence database pre-filtering heuristic to reduce the amount of time it takes to calculate matches against the HAMAP database (the heuristic is based on HMMER3.0, but the analysis still uses the core HAMAP algorithm, and is all implemented within the InterProScan software).  This again speeds up our protein match generation process and helps to safeguard against future data growth. The PIRSF and HAMAP databases were identified as being the slowest databases to calculate matches at at the start of 2014, but after work from both the database maintainers and the InterPro team, but this is no longer the case.

A leaner UniProtKB: 
At the same time, the number of proteins in UniProtKB has decreased significantly, where some 47 million sequences from highly redundant bacterial proteomes have been deleted (for details, see here, described half way down the page).

Faster and fitter InterPro production:
The majority of these developments have taken place under the hood, so it is unlikely that you will have been aware of our fitter and faster production system. What we hope you will notice, however, are more regular InterPro releases and more frequent member database updates in future, as these and other optimisations come into effect.

Alex Mitchell
on behalf of the InterPro team