Annual Scientific Report 2015

Pfam
Pfam is a database of protein sequence families. Each
Pfam family is represented by a statistical model (a
profile-hidden Markov model), trained using a curated
alignment of representative sequences. These models
can be searched against all protein sequences to find
occurrences of Pfam families, thereby aiding the
identification of evolutionarily related sequences. As
homologous proteins are more likely to share structural
and functional features, Pfam families can aid in the
annotation of uncharacterised sequences and guide
experimental work.
http://pfam.xfam.org
HMMER
HMMER is a sequence-analysis package that can be
used with both protein and nucleotide sequences. At
the core of the software is an algorithm that enables the
searching of one or more probabilistic models (profile
hidden Markov models, HMMs) against either a single
sequence or a database of sequences. The HMMER
website has implemented this software as a set of fast
web services, with both a programmatic interface
and graphical user interfaces. Profile HMMs are
incredibly powerful, allowing users to detect distant
evolutionary relationships.
www.ebi.ac.uk/Tools/Hmmer
Protein Function Development
Maria Martin
• Re-launched the Enzyme portal and developed new
interfaces and tools for UniProt and QuickGO, with a focus
on optimising user interaction with these websites;
• Implemented a method for identification of highly
redundant proteomes and removal from UniProtKB;
• Extended the provision of variants with consequences at the
protein level, incorporated variation data from ExAC and
the Exome Sequencing Project (ESP);
• Released experimental peptides mapped to UniProt proteins
from mass-spectrometry studies in collaboration with
PeptideAtlas and MaxQB;
• Extended the scope of the annotation tool Protein2GO and
the GO browser QuickGO, and implemented a PSIQUIC
server for protein-protein interaction annotations
visualisation in Cytoscape;
• Implemented the automatic annotation of domains, signal
peptides, transmembrane and coil-coil regions for millions of
protein sequences in UniProtKB/TrEMBL.
Protein Function Content
Claire O’Donovan
• In the context of the Consensus Coding Sequence (CCDS)
project, ensured the curated, complete synchronisation with
the HGNC, which has assigned unique gene symbols and
names to 39 000 human loci (19 001 of which are listed as
coding for proteins);
• Helped establish minimum standards for genome
annotation to enable scientists to exploit complete genome
and proteome datasets to their full potential;
• Improved UniProt Automatic Annotation by significantly
increasing the number of UniRules, with an emphasis on
enzymes across the taxonomic space;
• Secured funding to continue our contribution to the
validation of the computational approaches submitted to the
Critical Assessment of Function Annotation experiment.
Sequence Families
Rob Finn
• Refactored Pfam to utilise UniProt reference proteomes as
the underlying sequence database, streamlining curation
and production processes while minimising impact on
sensitivity;
• Optimised Pfam quality control to allow minor overlaps
between Pfam entries to allow better modeling of protein
families;
• Streamlined production and delivered monthly updates of
InterPro data to UniProt for their automatic annotation
procedures;
• Integrated a net gain of over 2000 new member database
signatures within InterPro, resulting in over 1800 new
entries;
• Provided GO terms to UniProt, with the latest release
assigning ~110 million terms to approximately 35 million
proteins in UniProt release 2016_01;
• Migrated the HMMER web services from Janelia Research
Campus;
• Expanded HMMER services to include PIRSF HMM
searches and support for UniProt reference proteomes, now
the default sequence database;
• Issued two releases of Pfam and six releases of InterPro.
2015 EMBL-EBI Annual Scientific Report 19