Knowledge Based Reactions

1
Knowledge Based Reactions

• Outside of quantum mechanical modeling, typical
  method is to explicitly specify the participating
  reactant atoms and how bonds are rearranged as a
  reaction profile

O
O
NH2-R2
H

H2O
OH
R1
NH2-R2
R1
Sample Reaction Profile
2
Knowledge Based Limitations

• Requires manual pre-specification of many
  different known reaction profiles to achieve any
  degree of generalization

3
Knowledge Based Limitations

• A very generic reaction profile can cover
  everything from previous slide and indeed about
  50 of all known organic reactions
• Still, a screening or ranking method is needed to
  filter many unrealistic reactions proposed above
• Many reactions with more sophisticated profiles
  are not covered without manually specifying more
  knowledge based profiles
• Diels-Alder
• Azide Alkyne aromatic cyclization

4
Reaction Favorability Scoring

• Simple scoring method to suggest predicted
  reaction feasibility and favorability based on a
  simplification of Hess law
• DHreaction S(BDEbroken) S(BDEformed)
• Estimate change in enthalpy by looking up
  bond-dissociation energies (BDE)
• Apply additional bonuses and penalties for
  aromaticity and ring strain

5
Pseudo-Mechanistic Reactions

• First step towards more generalized,
  pseudo-mechanistic reaction modeling is the
  introduction of intermediates
• Model breaking a bond in each reactant by
  separating charge, representative of the bond
  electrons moving to one atom
• Closing the intermediates is then just a matter
  of matching and - charges

A B C D
A B- C- D
A B C D
6
Pseudo-Mechanistic Reactions

• Applying general electron-shifting rules on the
  intermediates before closing them into products
  provides a lot of power and chemically intuitive
  results

7
Azide Alkyne Example
8
Diels-Alder Example
9
ChemDB Architecture
finger.ics.uci.edu
2. Similarity Query
1. HTTP Request
3. Similarity Results
cdb.ics.uci.edu
Client
4. DB Query
6. HTML Results
5. DB Results
chemdb.ics.uci.edu
10
ChemDB Architecture

• cdb.ics.uci.edu
• Linux, 1GB
• Primary Integration Point
• Web Server (Apache)
• Image Server (internal smi2gif.py script)
• Issue Tracker (Scarab, Tomcat JSP, PostgreSQL DB)
• chemdb.ics.uci.edu
• MS Windows Server 2003
• Database Server (PostgreSQL)
• finger.ics.uci.edu
• Linux, 4GB
• Similarity Search Server (Internal FINGER module)