Michel Dumontier , Ph.D. Assistant Professor of Bioinformatics

Accurate biochemical knowledge starting with precise structure-based

criteria for molecular identity

Michel Dumontier, Ph.D.Assistant Professor of Bioinformatics

Department of Biology, School of Computer ScienceInstitute of Biochemistry, Ottawa Institute of Systems Biology

Carleton University

01/04/20091 NCBO Seminar Series::Michel Dumontier

Problem Statement (I)

• Although biochemical events can be described with reference to specific chemical substances, we may want to describe them at finer/grainier levels of (mereological) granularity.– residue : post translational modification

– collection of residues : motif/domain/interaction site

– atom : atomic interactions, catalytic mechanism

– collection of atoms : binding/catalytic site, interaction

• This requires identifiers for parts, regions (contiguous and non-contiguous), aggregates/complexes.

• However, we do not (AFAIK) have a precise (reproducible) methodology to automatically generate these!


Bio2RDF: 2.3B triples of SPARQL-accessible linked biological data!

Chemical Parts!

Case Study: HIF1αHypoxia-Inducible Factor 1, alpha chain (uniprot:Q16665)

Master transcriptional regulator of the adaptive response to hypoxia

• Under normoxic conditions, HIF1α is hydroxylated on Pro-402

and Pro-564 in the oxygen-dependent degradation domain (ODD) by EGLN1/PHD1 and EGLN2/PHD2. EGLN3/PHD3 has also been shown to hydroxylate Pro-564. The hydroxylated prolines promote interaction with VHL, initiating rapid ubiquitination and subsequent proteasomal degradation.

Context Dependent Behaviora) Normoxic Conditionsb) Hypoxic Conditions

Multiple hydroxylations

Part of a domain

The part is the agent in the process

Selective interaction with parts


Are these the same?

• HIF1α – au naturel• HIF1α

– hydroxylated @P402

• HIF1α– hydroxylated @P564

• HIF1α– hydroxylated @P402 & @P564

• HIF1α– hydroxylated @P402 & (@P564)

– ubiquitinated @Lys-532

• HIF1α– L400A & L397A


NO!!!!

• These are structurally different• Each exhibits distinct functionality!

• Yet most databases (Uniprot/Genbank) don’t have separate identifiers for them

• Reactome has an internal identifier for referring to different forms, but links to Uniprot entries and doesn’t provide an explicit description of the structure that it corresponds to!


So

We have a clear need for being able to refer to distinct biochemical entities, based at least on their structure.

We also need to refer to arbitrary structural parts.

Should we generate all the combinations a priori???

NO!!

Should we be able to automatically generate the identifier from the structural attributes?

-> YES!!!

Should we semantically annotate (manually or otherwise) those forms known to be involved in specific processes???

-> YES!!!

What identifiers are unique for a given structure?


InChI• IUPAC International Chemical Identifier (InChI)• A data string that provides

1. the structure of a chemical compound

2. the convention for drawing the structure

• Different compounds must have different identifiers. Several attributes can be used to distinguish one compound from another. – chemical graph (connection table) – Formula– Atom type (only some atoms explicit)– Bond type– Stereochemistry– Mobile/fixed H-bonds (tautomers)– Isotopic composition– Atomic charge


(S)-Glutamic Acid

InChI={version}1/{formula}C5H9NO4/c{connections}6-3(5(9)10)1-2-4(7)8/h{H_atoms}3H,1-2,6H2,(H,7,8)(H,9,10)/p{protons}+1/t{stereo:sp3}3-/m{stereo:sp3:inverted}0/s{stereo:type (1=abs, 2=rel, 3=rac)}1/i{isotopic:atoms}4+1


AuxInfo={version}1/{normalization_type}1/N:{original_atom_numbers}5,6,2,7,1,4,8,9,10,11/E:{atom_equivalence}(7,8)(9,10)/it:{abs_stereo_inverted:sp3}im/I:{isotopic:original_atom_numbers}/E:{isotopic:atom_equivalence}m/rA:{reversibility:atoms}11nCCHN+CCC.i13OOOO/rB:{reversibility:bonds}s1;N2;P2;s2;s5;s6;s7;d7;d1;s1;/rC:{reversibility:xyz}6.1671,-19.3365,0;7.0125,-18.4864,0;6.4113,-17.4485,0;7.6089,-17.4485,0;7.8578,-19.3318,0;8.891,-18.7306,0;9.7363,-19.576,0;9.7316,-20.7735,0;10.8916,-19.266,0;5.0071,-19.0265,0;6.1624,-20.534,0;

AuxInfo=1/1/N:5,6,2,7,1,4,8,9,10,11/E:(7,8)(9,10)/it:im/I:/E:m/rA:11nCCHN+CCC.i13OOOO/rB:s1;N2;P2;s2;s5;s6;s7;d7;d1;s1;/rC:6.1671,-19.3365,0;7.0125,-18.4864,0;6.4113,-17.4485,0;7.6089,-17.4485,0;7.8578,-19.3318,0;8.891,-18.7306,0;9.7363,-19.576,0;9.7316,-20.7735,0;10.8916,-19.266,0;5.0071,-19.0265,0;6.1624,-20.534,0;

More non-core info captured in “AuxInfo” string...


So... InChi a really just a cryptic data identifier

Clever software required to gradually build the chemical identifiers in a series of well-defined steps –

normalization, canonicalization then serialization

Humans can’t (easily) generate them nor can they easily understand them. But that’s OK.

It’s not (user) extensible. But that’s OK.


• Possible... but a 1000 residue protein would contain ~15,000 atoms on average.... – OpenBabel seemed to struggle with anything over 100

residues • Maybe needs some performance tweaking?

– Size of the string will be enormous• We can use InChiKeys (SHA1 hash), but then we need to provide a

you-submit-InChI, we-store-both and they-look-it-up service.

– Modularize InChI construction for (linear) polymers?• Make InChi strings for each residue, and concatenate – rename the

atoms according to the residue position

– We still need to translate the InChi string ...

InCHI for Proteins???


CMLSDF

O1[C@@H]([C@@H](O)([C@H](O)([C@@H](O)([C@@H]1(O)))))(CO) 79025

IUPAC

InChI=1/C6H12O6/c7-1-2-3(8)4(9)5(10)6(11)12-2/h2-11H,1H2/t2-,3-,4+,5-,6+/m1/s1InCHI

α-D-Glucose

6-(hydroxymethyl)oxane-2,3,4,5-tetrol OR (2R,3R,4S,5R,6R)-6 -(hydroxymethyl)tetrahydro -2H-pyran-2,3,4,5-tetraol

SMILES

OpenBabel

OWL Has Explicit Semantics

Can therefore be used to capture knowledge in a machine understandable way


Chemical Ontology

Chemical Knowledge for the Semantic Web.Mykola Konyk, Alexander De Leon, and Michel Dumontier. LNBI. 2008. 5109:169-176. Data Integration in the Life Sciences (DILS2008). Evry. France.

http://134.117.55.46/pubs/publications.html#Mykola%20Konyk

http://134.117.55.46/pubs/publications.html#Alexander%20De%20Leon

http://134.117.55.46/pubs/publications.html#Michel%20Dumontier

http://dils2008.lri.fr/

http://code.google.com/p/semanticwebopenbabel/


hydroxyl groupmethyl group

Knowledge of functional groups is important in chemical synthesis, pharmaceutical design and lead optimization.

Functional groups describe chemical reactivity in terms of atoms and their connectivity, and exhibits characteristic chemical behavior when present in a compound.

Describing chemical functional groups in OWL-DL for the classification of chemical compounds

N Villanueva-Rosales, MDumontier. 2007. OWLED, Innsbruck, Austria.

Ethanol


Describing Functional Groups in DL

HydroxylGroup: CarbonGroup that (hasSingleBondWith some (OxygenAtom that hasSingleBondWith some HydrogenAtom)

OHR

R group


Fully Classified Ontology

35 FG


And, we define certain compounds

Alcohol: OrganicCompound that (hasPart some HydroxylGroup)


Organic Compound Ontology

28 OC


Question Answering

Query PubChem, DrugBank and dbPedia*

* Requires import of relevant URIs01/04/200922 NCBO Seminar Series::Michel Dumontier

But...

• Molecules represented as individuals because OWL-DL only allows tree-like class descriptions– No variable binding (e.g. ?x) ... no cyclic

molecule/functional group descriptions at the class level

• Boris Motik et al has a proposal for Description Graphs, – Robert Stevens & Duncan Hull trying it out for

chemical representation....


Identifiers for Atoms

• Atom identifiers can be consistently retrieved from the OpenBabel model.– Canonical numbering means we can reliably refer to a

specific region rather than a (possibly degenerate) sub-graph match.

– In our plugin, URI component naming was based on the assigned molecule identifier

e.g. pubchemid#aN, where N is the number

– Use InChiKey as base?e.g. InChiKey#aN


What about identifiers for collection of atoms?

• Potentially useful in describing residues, PTMs, binding sites, etc. – Is the lack of connectivity sufficient?

• Contiguous: – ranges (aN-aN)– enumerations (aN,aN,aN)

• Non-contiguous:– Combination of ranges, enumerations?


Can we reuse our positional nomenclature for residues?

• Residues are generally referred to by their absolute position in the biopolymer sequence.

e.g. Pro @ X on Protein Y

InChiKey#a50-a65 owl:sameAs InChiKey#r5

InChiKey#r5_a1-r5_a15 owl:sameAs InChiKey#r5

• Collection of Residues might follow the same rules as a Collection of Atoms.– Useful for defining domains, motifs, etc


• We already have a simplified representation for biopolymers... – Canonical sequence is represented by a string of

single letter characters• DNA: ACGT• RNA: ACGU• Proteins: 20 amino acids (not B,J,O,U,X,Z)

– Modifications can be referred to with ChEBI/PSI-MOD ontology (e.g. Prolyl hydroxylated residue @ 402)

• Each (modified) residue must have its InChi description so as to capture explicit structural deviations (de-protonation, etc)

An Alternative Scheme


PSI-MOD contains modified residues with links to structural descriptions


But what if we have a modification that isn’t contained in the ontology!

• No problem... define your own term, with the corresponding structural description (InChi, SMILES), and add to an ontology document...– If you’re using OWL, you can add the import

statement and publish it.

• And, of course, you should submit it to the appropriate ontology development teams. (and later make it equivalent to)


While we’re at it, we could extend our expressive capability to match that of OWL:

• Specification – Exactly mod1@pos X– Only mod1@posX

• Minimum : – At least mod1@posX

• Combination:– mod1@posX AND mod2@posY, X != Y

• Possibilities/Uncertainty: – (mod1 OR mod2) @posX

• Exclusion:– not mod1 @ posX


So what if...we describe the structural features of the molecule with OWL (sequence + PTMs), and generate an identifier from one of its serializations (RDF/XML?)

that way we have the explicit description as the identifier in a form that is compatible with the semantic web.



Uniprot example revisitedUnder normoxic conditions, HIF1α is hydroxylated on Pro-402

and Pro-564 in the oxygen-dependent degradation domain (ODD) by EGLN1/PHD1 and EGLN2/PHD2. The hydroxylated prolines promote interaction with VHL, initiating rapid ubiquitination and subsequent proteasomal degradation

.

:A rdfs:subClassOf :Hydroxylation:A hasParticipant (:0#r402 and :Substrate):A hasParticipant (:1#r402 and :Product):A hasParticipant (:5 and :Enzyme)

:B rdfs:subClassOf :Interaction:B :hasParticipant (:2#r402 or :3#r564 or :4#r402,r564):B :hasParticipant (:6)

:1 (HIF1α):2 (HIF1α + P402hyd):3 (HIF1α + P564hyd):4 (HIF1α + P402hyd + P564hyd):5 (EGLN1):6 (VHL)

Please ignore the made up short-hand syntax!


Infering Protein Participation

• OWL Role Chain

hasParticipant o isPartOf -> hasParticipant

if process has the part as a participant, then the whole is also a participant

:0#r402 :isPartOf :0:1#r402 :isPartOf :1

:A rdfs:subClassOf :Hydroxylation:A hasParticipant (:0#r402 and :Substrate):A hasParticipant (:1#r402 and :Product)

:A hasParticipant :0:A hasParticipant :1


Contextual, but non-structural considerations in identifier generation?

• Chemical?– pH?– Temperature?– Environment (in vitro, in vivo, in silico)?

• Biological?– Species?– mRNA/Gene from which it was transcribed/encoded?

• Indirect Relationships?– Point & Multiple Mutations?– Alternative Splice Variants?– Sequence Similarity?


Summary

• We need a precise method to generate identifiers for biopolymers and arbitrary sets of their parts.

• Consistent identifier generation will allow anybody to specify findings according to the biopolymers for which it was observed, whether it exists in a database or not, and will allow us to link biochemical knowledge at finer levels of granularity.

• (at least) two identifier schemes were put forward to initiate discussion, with the goal of setting a standard naming convention.


dumontierlab.com

[email protected]

Special thanks to PhD Student Leonid Chepelev for insightful discussions

semanticscience.org01/04/200937 NCBO Seminar Series::Michel Dumontier

Documents

Michel Dumontier , Ph.D. Assistant Professor of Bioinformatics