?

Functional Site rule: tags active

site, binding, other residue-specific information

Functional Annotation rule: gives name, EC, other activity-specific information

www.uniprot.org

Swiss Institute of Bioinformatics (SIB)

European Bioinformatics Institute (EMBL-EBI)

Protein Information Resource (PIR)Contact help@uniprot.org

UniProt is mainly supported by the National Institutes of Health (NIH) grant 2 U01 HG02712-04. Additional support for the EBI's involvement in UniProt comes from the European Commission contract FELICS (021902) and from the NIH grant 5 P41 HG02273-06. UniProtKB/Swiss-Prot activities at the SIB are supported by the Swiss Federal Government through the Federal Office of Education and Science. PIR activities are also supported by the NIH grants for NIAID proteomic resource (HHSN266200400061C) and grid enablement (NCI-caBIG-ICR), and National Science Foundation grants for protein ontology (ITR-0205470) and BioTagger (IIS-0430743).

The PIRSF Protein Classification System as a Basis for

Automated UniProtKB Protein Annotationpir.georgetown.edu/pirsf

Correct protein annotation relies on both global (whole protein) and local (domain and motif) sequence similarities. We have developed a method by which annotation of site-specific features can be confidently propagated from experimentally-characterized proteins to uncharacterized proteins. The method relies upon rules that identify the specific amino acids in a protein chain eligible for tagging with appropriate information. Rules are specific for a particular protein family, and rely upon the identification of active site, binding site, modified or other functionally-important residues in a template sequence.

A general approach for functional characterization of an unknown protein is to infer function based on similarity to a “best-hit” protein in sequence databases. This is powerful method is nonetheless susceptible to error. In part, these errors can be avoided by using a curated, hierarchical, whole-protein classification database.

The advantage is conferred by “strength in numbers.” Instead of relying on the (hopefully) accurate annotation of a single (hopefully related) protein (usually, the BLAST best hit), using curated classification databases allows reliance on the collected wisdom of multiple proteins, or at least the assurance that the members are truly related.

The annotation power of protein classification databases is even more powerful if a single database contains families with progressively greater levels of similarity (that is, hierarchies). Theoretically, one query could be confidently predicted to be a member of a parent family, but not a child family, while a different query might be confidently assigned to both levels. In such cases, the most-specific possible annotation could be propagated.

Despite these precautions, use of protein classification databases cannot resolve one particular source of error: asserting that a query protein has a particular enzymatic activity, even though it lacks the specific residues responsible for that activity. This is because all current classification algorithms rely on global similarity to make functional inferences. Here we describe a more robust method—PIR Site Rules—for inferring function of uncharacterized proteins. The method has the added benefit of being able to explicitly flag the residues important for a given activity (feature [FT] line), and allowing a rule-based annotation of other data fields, such as protein name (definition [DE] line). After a brief summary, the case of ATP- and PPi- dependent phosphofructokinases will be presented.

Summary• Position-Specific Features:

– active sites

– binding sites

– modified amino acids

• Current requirements:– at least one PDB structure

– experimental data on functional sites

• Rule Definition:– Select template and align with

PIRSF seed members

– Edit MSA: retain conserved regions covering all site residues

– Build Site HMM from concatenated conserved regions

Algorithm• Match Rule Conditions

– Membership Check• Ensures that the annotation is

appropriate

– Conserved Region Check (site HMM threshold)

– Site Residue Check (all position-specific residues in HMMAlign)

• Propagate Information– Feature annotation using

controlled vocabulary

– Evidence attribution (experimental/computational)

– Attribute sources and strengths of evidence

PIRSR000532-4

PIRSR000532-5

Figure 1. Annotate carefully. Annotation is propagated only if all the required residues are present. A protein (P29780) fails the alignment test (red oval), since the rule calls for a cysteine at that position (the query has a glycine). The other two residues (green ovals) are a match. Nonetheless, no information will be propagated (red arrow).

The Phosphofructokinase (PFK) Case:

ATP versus PPi dependence

Figure 2. Importance of annotation based on site-specific features. Rule definition begins with knowledge about residues important for catalytic activity or binding. PFK is a key regulatory enzyme in the Embden-Meyerhoff glycolytic pathway. Classification of PFK proteins revealed that major functional specialization can occur as a result of even a single amino-acid residue change. Two amino acid positions (105 and 125, E. coli numbering) are critical determinants of ATP or PPi utilization (boxes), with one position especially key (arrow). The ability to use ATP depends on the presence of a glycine at the indicated position. Accurate propagation of protein function therefore depends on crafting rules that take advantage of the ability to distinguish between these possibilities, as illustrated in the next figure.

Figure 3. PIR Site Rule (PIRSR) definition. Important residues on a template sequence are indicated, along with the appropriate annotation if a query passes all match conditions. Two rules regulate the annotation of ligand. A match to rule 4 means the query is ATP-dependent, while a match to rule 5 means the query is PPi-dependent.

Figure 4. Global similarity check. A Leifsonia protein was tested against HMMs for protein families. Q6AG22 matches PIRSF000532, a family that contains mostly ATP-PFKs, but also a few PPi-PFKs.

Figure 5. Further confirmation. All the proteins hit by Q6AG22 using BLAST are members of PIRSF000532 (only partial results are shown), hence the protein was added to this family. Note that the best characterized matches are PPi-PFKs, but….

PPi-PFK

ATP-PFK

Figure 6. Looks can be deceiving. Iterative clustering using BlastClust makes the initial observation that the query (red arrow) might be a PPi-PFK (blue arrows) less certain.

Figure 8. Functional variation within one protein family: binding sites with different specificity drive choice of applicable rule to ensure appropriate annotation. Members of the phosphofructokinase (PFK) family evolved into ATP- or PPi-dependent forms. While propagating the name “Phosphofructokinase” to all members would not be inaccurate, it fails to take full advantage of current knowledge. The residues that contribute to each dependency are known (see Functional Site rules PIRSR000532-4 and PIRSR000532-5). Therefore, DE line annotation should depend on which site rule “fits,” if any. The schematic above indicates the tests and results that occur to propagate name information. Members of PIRSF000532 are tested against the relevant site rules (black arrows). A positive result for the ATP-dependency rule (green arrow) means that the entry would be named “ATP-dependent phosphofructokinase,” according to the PIR Name Rule (PIRNR) PIRNR000532-1, while a positive result for the PPi-dependency rule (blue arrow) means that the entry would be named “Pyrophosphate-dependent phosphofructokinase” (PIRNR000532-2). Note that failure to match either one does not mean the activity is missing. Thus, a fall-back rule can be created (the “zero rule,” dotted red arrow) that would propagate simply “Phosphofructokinase” without any qualifier. In the case of the query Q6AG22, having failed both specificity rules, the zero rule would apply.

Another possibility. Most ATP-PFKs have the G-G combination, while most PPi-PFK have the D-K combination. Hence, only rules for these two possibilities were written. However, recent evidence indicates that the G-K combination also functions as an ATP_PFK. Thus, a new rule can be written to cover this scenario, and accordingly Q6AG22 would be annotated as an ATP_PFK.

ConclusionCritical to our understanding of biology is accurate and up-to-date information. The process of evolution affords us the ability to make inferences about the nature of the proteins that govern biological processes, since like proteins often perform like (if not exact) functions. Unfortunately, this same process has been far from a smooth transition from state to state. The result is that inferences made about one protein based on similarity to another protein using automated methods are often suspect. This is more than a mere annoyance. The lack of rigorous methods for propagating appropriate information hampers knowledge discovery by either reducing the associations that can be made, or by producing associations that should not be made. However, the recent development of methods for better annotation hold much promise for preventing—and even reversing—the previous trend toward rampant misinformation. The combination of hierarchical, whole-protein classifications and rule-based large-scale annotation pipelines is a significant step in the right direction

Q9KH71 DALIAIGGEDTLGVASKFSKLGLPMIGVPKTIDKD

query DAIIAIGGEGTLTAARRLTDAGLRIVGVPKTIDND

P0A796 DALVVIGGDGSYMGAMRLTEMGFPCIGLPGTIDND **::.***:.: * :::. *: :*:* ***:*

PIRSR000532-5 comparison

PIRSR000532-4 comparisonFAIL

FAIL

Figure 7. Motif check. The query protein was tested against each of the rules governing ligand binding. Rule 4 (bottom) stipulates that annotation of ATP dependence requires a G-G combination in key positions. The query passes for only the first position, and thus fails the test. Rule 5 (top) stipulates that annotation of PPi dependence requires a D-K combination in key positions. The query passes for only the second position, and thus fails the test.

What to do?

Site Rules Status

• 301 PIR site rules covering 168 PIRSFs have been defined.• Site information was imported from the Catalytic site residue dataset and

Catalytic Site Atlas.• 32 PIR site rules covering 19 PIRSF families have been manually curated

and submitted to SIB for comments & suggestions. • The SIB suggestions will be incorporated and logfiles for annotation

propagation to Swiss-Prot entries not already in HAMAP will be submitted to SIB.

D.A. Natale, C.R. Vinayaka, and C.H. Wu. Large-scale, classification-driven, rule-based functional annotation of proteins. Encyclopedia of Genetics, Genomics, Proteomics and Bioinformatics. Bioinformatics Volume, Subramaniam, S. (Ed.) John Wiley & Sons, Ltd. 2004.