TCM -- Traditional Chinese Medicines EuroMUG 2000, Dave Weininger, Daylight CIS

Why TCM?

• Intrinsic value
  • 3000 years of clinical trials on 1,000,000,000+ subjects
  • almost all WAM pharmacological classes are represented

• Complex data model
  • raw data are relationships between entities
  • knowledge represented by relationships of relationships
  • handled poorly by current database management systems
  • information problem is similar/complementary to QSAR

TCM vs. WAM

• Differences
  • TCM system/qi vs WAM reductionist/systems
  • TCM herbal combination preps vs WAM synthetic pure substances
  • TCM acupuncture, WAM surgery
  • TCM old vs. WAM young

• Similarities
  • profound emphasis on use of potent drugs
  • formalized knowledge base, training
  • formalized drug preparation protocol

Why now? Two bridges.

• Molecular structures of drugs
  • Molecular structures now available for both pharmacopia's
  • TCM the book; WAM: WDI, medchem, etc.

• Diseases
  • Oriental & Western patients are 99.99% identical
  • bases of medical traditions are also nearly identical
    • aging process
    • reproductive problems
    • response to infection and toxins
    • psychological/psychiatric problems
    • trauma, military medicine
  • differences are mainly sociological and historical

TCM data sources

• Long history
  • 2000-3000 years of oral tradition
  • major revision about 1000 years ago
  • medical text appeared within 7 years of 1st printing
  • very insular until recently
  • almost all of it is "encoded" in Mandarin

• Current data sources

  • Traditional Chinese Medicine, the book.
    • book is available
    • authors X. Yan, Z. Zhou and G. Xie are available and helpful
    • published by Ashgate, ed. G. W. A. Milne is available and helpful
    • good internal computerized data format
    • uses DCM terminology (in theory)

  • A Practical Dictionary of Chinese Medicine, Weissman, et. al.
    • published by Paradigm; book, authors and editor are available
    • excellent, careful, respectful work
    • difficult electronic format CTeX, a LaTeX variant
    • BIG-5 and pinyan identifiers, needs modernization

TCM data

• Indications and effects
  • indexed by TCM "medicine name"
  • almost always framed as TCM concept
  • occasionally linked to WAM concept (<20%)

• TCM medicine
  • unique name is theoretically rigorous, but not so in practice
  • often composed of many components
  • unwritten rules of component equivalence (e.g., BA PO)

• TCM source
  • source of medicine component
  • often a plant, but not always
  • name is often synonymous/homonyminous with medicine
  • Latin taxonomic identifier available at family level
  • biological similarity metric

• TCM chemical
  • indexed by source
  • molecular structure really is a universal language!
  • molecular similarity metric for substance and mixture

TCM data relationships

• Indications and effects

  • to self:
    • similarity and membership are common
    • identity is rare

  • to WAM indications, usage, effects:
    • membership is common
    • similarity and identity are rare

  • to TCM medicine:
    • nearly complete mapping
    • almost always consistent

  • to TCM structure:
    • many-to-many mapping
    • similarity is significant but untested

• TCM medicine

  • to self:
    • similarity and membership are common
    • never identity (?)

  • to WAM drugs:
    • membership via indications/effects is rare (probably due to uncontrolled language)
    • membership via structure is common
    • significance is uncertain

  • to TCM source:
    • nearly complete mapping
    • suspect that mapping is incomplete
    • needs work (which we know how to do)!

  • to TCM structure:
    • nearly complete mapping
    • coverage is irregular

• TCM source

  • to self:
    • structural similarity is common
    • taxonomic similarity is common
    • I&E similarity is moderate
    • identity is rare but not zero :-(

  • to WAM sources:
    • untested
    • are such data available in a WAM natural products database?

  • to TCM structure:
    • nearly complete mapping
    • coverage is irregular

• TCM structures

  • to self:
    • structural similarity is common (highly clustered)
    • taxonomic similarity is common (source-correlated)
    • diversity is high (but not necessarily of "actives")

  • to WAM structures:
    • structural identity is moderate (some stereo problems)
    • structural similarity is high (TCM includes 80+% WAM)
    • cross-clustering not done yet :-(

TCM interface

• THOR database model is suitable

  All relationships are representable (though not terribly efficiently)

• Merlin search engine is not needed

  Small number of structures; can be handled by client interface "on-the-fly".

• Conventional interfaces are profoundly inappropriate

  The most important data are relationships which are lost when data model is flattened.

• Web-surfing interface appears to be suitable

  A web-like interface provides virtually all of the required functionality. In this model, the TCM database looks like a heavily cross-linked website. Similarity is the fundamental relationship. Data is extended by results of multivariate methods such as clustering and discriminant analysis. Users "surf" in this dataspace.

• Demo

Preliminary conclusions

• Rigorous data extraction is possible

  more work is needed

• TCM can be produced as a conventional database

  THOR database format is suitable

• A specialized interface will be required

  Simple forms and spreadsheets are inadequate.

Thanks to

• Ashgate

• Paradigm

• Home