Transcript Classifying chemistry: Current efforts in Canada

Classifying Chemistry: Current
Efforts in Canada
Chemistry, Data & the Semantic Web
251st ACS Chemistry Meeting, San Diego CA
David Wishart
University of Alberta
March 15, 2016
Chemical Classification –
Why Do It?
Zoologists Do It
Chemical Classification –
Why Do It?
Geologists Do It
Chemical Classification –
Why Do It?
Astronomers Do It
Chemical Classification –
Why Do It?
Druggists Do It
Chemical Classification –
Why Do It?
Lipid Chemists Do It
Chemical Classification
• Chemists started doing it (the periodic
table of the elements)
• Chemists led the way with standardized
naming (IUPAC nomenclature)
• Chemists led the way with hash keys
and standardized identifiers (InChI)
• But chemists are now far behind other
fields with respect to compound
classification
Chemical Classification Benefits
• Gives “order” to a complex field
• Forces clear and near-universal
definitions
• Establishes an ontology (a vocabulary
of terms, concepts and relationships)
• Improves searching and comparisons
• Helps identify relationships or origins
• Allows for automated annotations
Chemical Classification – A Start
• PubMed + MeSH (Biomedical only)
– 8319 compound related MeSH terms/classes
• ChEBI (Biological interest only)
– 320 chemical groups, 187 applications, 113
biological roles, 82 chemical roles
• Lipid Maps (Lipids only)
– 522 lipid classes
• OntoChem (More general)
– 7916 compound concepts, 32,469 synonyms
Why We Got Into It
Why We Got Into It
• Started manual classification for HMDB
and DrugBank in 2005 (partial)
• Growing number of requests from
metabolomics researchers wanting
compounds clustered by chemical type
• Challenges with manually annotating
multiple chemical/metabolite databases
with large numbers (100,000+) of
different compounds
How To Classify?
• By chemical or biological origin?
– Not always clear, requires manual annotation, not
many classification categories
• By function, application, chemical role?
– Requires considerable manual effort, always
changing, lots of categories, needs ontologies
• By biological pathways or biological role?
– Only works for biological compounds, not for
industrial or synthetic compounds, lots of unknowns
• By structure?
– Lots of categories, existing classifications to build on,
potentially automatable, works for all chemicals
Classifying by Structure
• Need to respect previous classification
schemes (amino acids, lipids, etc.) even if
they are not “purely” structural
• Need to find a preferred nomenclature (many
names can exist for the same compound
class)
• Need clear definitions and well-defined
categories
• Need to handle hybrid or chimeric structures
consistently and logically
The ClassyFire Server
• A webserver (and
database) designed to
facilitate chemical
classification and
chemical description
via structure alone
• Accepts InChI or
SMILES strings and
generates
classification in <0.5 s
http://classyfire.wishartlab.com
ClassyFire Schema
ClassyFire Features
• Spans the chemical space from natural products to
polymers, biomimmetics, inorganic & organic chemicals
• Fully automated
• Consistent and aligned with manual classification
schemes in MeSH (PubChem), ChEBI and Lipid Maps
• Every compound class is fully defined via a text
description and a computable structure definition
• Provides consistent naming system for chemical classes
• Includes 4822 inorganic and organic compound classes
• Has been used to classify all chemicals in PubChem,
ChEBI, KEGG, Lipid Maps, DrugBank, HMDB, YMDB, etc.
All Public Chemicals Are Now Classified
DrugBank Example
Classification - What’s Next?
• Feature Attributes
–
–
–
–
Name/InChI
2D/3D structure
Structural taxonomy
Similarity
• Physical Attributes
– QSAR descriptors
– PhysChem properties
– Organoleptic qualities
The 4 “F’s”
• Functional Attributes
–
–
–
–
–
–
Hazard properties
Health effects
Disease associations
Biological role
Industrial role
Membership list
• Fate Attributes
– Biological location
– Origin
– Pollutant status/fate
How To Get These Data?
• Calculate chemical properties using
open access or commercial tools
• Automatically extract information or
labels from existing public databases
• Use software to cross-check data
consistencies
• Perform text mining to extract detailed
relationships and/or knowledge
• Do it manually
The Tools
DataWrangler
• Automated tool to calculate, extract and
verify structure, InChI, names,
synonyms, formula, MW, chemical
properties, descriptions (local or
Wikipedia), chemical classification
(ClassyFire), pathways, targets,
reactions and “some” ontological terms
• Accepts CAS, InChI, SMILES or name
PolySearch 2.0
• Online text-mining system for identifying
relationships between human diseases,
genes, proteins, drugs, metabolites, toxins,
metabolic pathways, organs, tissues,
subcellular organelles, positive health
effects, negative health effects, drug actions,
Gene Ontology terms, MeSH terms, ICD-10
medical codes, biological taxonomies and
chemical taxonomies
• Supports generalized 'Given X, find all
associated Ys' query, where X and Y can be
selected from the above biomedical entities
PolySearch 2.0
www.polysearch.ca
PolySearch 2.0
• Maintain a local API for “heavy” queries
• Searches local versions of Wikipedia,
PMC-Central, PubMed, NCBI On-line
textbooks, US Patent abstracts,
UniProt, DrugBank, HMDB, T3DB,
ECMDB, YMDB, DailyMed, KEGG,
OMIM, HPRD, MetaCyc, NCBI taxonomy
• 165 Gbytes of data
ChemoSummarizer
(R)-Pabulenol
General Characteristics
Function or Role
• Enter an InChI code or a
SMILES string and a 5001100 wikipedia-like
description is
automatically generated
(with references +
pictures)
• Intended to help
automate compound
descriptions
• Plans to put the data into
a machine readable
format (RDF)
Looking Forward
• Consistent and complete mapping
between MeSH, Lipid Maps, ChEBI,
OntoChem and Classyfire – a common
chemistry taxonomy
• A common, comprehensive chemistry
ontology (the 4 “F’s”)
• Common data formats and exchange
protocols between major chemistry
databases (RDF)
Looking Forward
• New tools that use ontologies to
generate novel information and novel
hypotheses
• Tools that extract previously unknown
or little-known information from the
literature
• Creation of automated, open access
chem-ontology servers like AMIGO (for
gene ontology) – call it AMICO?
Looking Forward
•
•
•
•
•
•
•
•
Filling in missing data on known cmpds
Predict Likely Biological Function
Predict Likely Industrial Role
Predict Likely Source Organisms
Predict Likely Toxicity or Hazard
Predict Likely Health Effects
Predict Likely Pathways/Targets
Predict Likely Organoleptic Properties
Looking Forward
• Analyzing Newly Synthesized or Newly
Discovered Compounds
• Predict Likely Biological Function
• Predict Likely Industrial Role
• Predict Likely Source Organisms
• Predict Likely Toxicity or Hazard
• Predict Likely Health Effects
• Predict Likely Pathways/Targets
• Predict Likely Organoleptic Properties
Acknowledgements
•
•
•
•
•
•
•
Yannick Djoumbou
Tanvir Sajed
Yifeng Lu
Zachery Budinsky
David Arndt
Craig Knox
Michael Wilson