2. Molecular Representations

1
2. Molecular Representations
2
Communicating Chemical Data

• Chemical data Text, numbers, and molecules
• Standard valence model of chemistry
• Discrete bonds represent shared electrons
• Codify into a reproducible representation
• Graph of atoms and
• bonds is most commonly
• understood representation

3
2D Graph of Atoms / Bonds

• Labeled graph
• Nodes Atoms, Symbols C, N, O, H,
• Edges Bonds, Order 1, 2, 3, aromatic,..
• Organic compound shorthand
• Assumed carbons
• Implicit hydrogens
• Standard valence rules

4
Tractability

• Small, Tree-Like, Graphs
• Number of vertices is small (e.g. less than 50)
• Number of edges is small (average degree 2.3 or
  so)
• Tree-like

5
2D Data Formats

• Bond matrix formats
• exist but size nAtoms2
• Connection table
• List of nodes
• C1, C2, O3, N4
• List of edges
• 1-2, 23, 2-4
• SDFile, Mol2 Formats
• Not human writeable

6
1D Line Notations

• Should be human parseable to facilitate
  communication without computer module
• Nomenclature
• IUPAC system 2-amino-3-phenyl-propanoic acid
• Common names phenylalanine
• SMILES C(C(O)O)(Cc1ccccc1)N
• Widely used, non-standardized
• InChi
• Recent, IUPAC supported official standard
• Ex. 1/C9H11NO2/c10-8(9(11)12)6-7-4-2-1-3-5-7/h1-5,
  8H,6,10H2,(H,11,12)

http//www.iupac.org/inchi/
7
IUPAC Nomenclature

• IUPAC Standard Naming Conventions
• propane
• propanoic acid
• 3-hydroxy-propanoic acid
• 2-amino-3-hydroxy-propanoic acid
• Unwieldy standard and inconsistent adoption
• Common names and abbreviations (Serine)
• Systematic bidirectional translation unreliable

8
SMILES Basics

• Connection tables as a character string
• Atoms Atomic symbols C,N,O,S,
• Bonds single - (implicit), double , triple
  
• Examples
• CBr
• CO
• CN
• OCCN

http//www.daylight.com/smiles/
9
SMILES Basics

• Branching Parentheses
• Cycles Numerical annotations
• CCC(O)C
• CC(N)(N)O
• C1CCCC1
• N12CCCCC1CCCC2
• NCC(CN)N1CCCC1
• Extensions for
• Inorganic atoms, unusual valence, formal charges,
  stereochemistry, aromaticity, reactions, etc.

10
Canonical Representations

• Unique representation needed for rapid DB lookup
  and to check uniqueness
• Need to uniquely order the atoms of a molecule
• nAtoms! atom orderings possible
• Morgan Algorithm
• Label nodes by connectivity (heavy degree)
• Relax iteratively towards extended connectivity
  (EC) using neighbor values
• Use EC magnitude to decide on atom order
• EC tie-breaking by atom, bond distinctions

11
Stereochemistry / Isomers

• Chemical handedness
• Same connectivity, but not superimposable
• Atoms with at least 4 distinct components
• Double bonds with distinct components at ends
• Specification by atom / bond labels

e.g. CC_at_H(N)O vs. CC_at__at_H(N)O
e.g. O/CC/N vs. O/CC\N
12
3D Atomic Coordinates

• 2D graph only specifies connections
• 3D spatial coordinates (center, radius, surface)
• Largely unavailable
• Usually predicted

13
4D Conformers

• Molecules are relatively rigid w.r.t.
• Bond length
• Bond angles
• Single bonds are very
• flexible w.r.t rotation
• More information with
• collection of multiple,
• static 3D conformations

14
Molecular Surfaces

• For intermolecular interactions, externally
  visible surface is most important
• Representations Orbitals,VDW Radii,
  Accessibility, Tessellations

http//www.netsci.org/Science/Compchem/feature14e.
html
15
Valence Model Limitations

• Bonds are non-existent
• Model of shared electron orbitals
• Difficulty modeling
• Aromaticity
• Resonance
• Tautomers
• Etc.

16
Structural Keys

• Motivated in part by rapid screening for
  functional group substructures
• Pre-compute presence of common / important
  substructures up front and record in bit-vector
• Example of structural keys
• Presence of atoms (C, N, O, S, Cl, Br, etc.)
• Ring systems
• Functional Groups
• Aromatic, Phenol, Alcohol (ROH),
• Amine (RNH2), Acid(RC(O)OH), Ester,

17
SMILES Examples
18
SMILES Examples
19
SMILES Examples
20
(No Transcript)
21
Generalized Fingerprints

• Structural Keys
• Generalizes only in proportion to knowledge
• Sparsely populated
• Good screening filter will have thousands of
  keys, but each item generally only has a few
  dozen
• Generalized Fingerprints (Spectral
  Representations)
• No pre-defined patterns
• Record counts or presence/absence of
  substructures (e.g. labeled paths, trees, etc)
• Fixed length (binary) vectors
• Fast algorithms
• Abstract, hard to traceback meaning of individual
  bits

22
Systematic Graph Features

• For chemical compounds
• atom/node labels
• A C,N,O,H,
• bond/edge labels
• B s, d, t, ar,
• Trace Paths
• Depth First Search

(CsNsCdO)
23
Fingerprint Flowchart

• 0 Bonds
• O
• C
• N
• 1 Bond
• OC
• CC
• CN
• 2 Bonds
• OC-C
• C-C-N
• 3 Bonds
• OC-C-N

Graph Feature Extractor
Random Number Generator
Hash Function
Modulo FP Size
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 0 1 0 0 0 0 1 1 1 0 1 0 1 0 0
24
Other Fingerprint Representations

• Derived Representations
• Information Compression
• Example Local Sensitive Hashing (LSH)
• Choose K random lines in high-dimensional space
• Project data points
• Bin coordinates

25
Summary

• Rich set of representations
• 1D SMILES, Fingerprints
• 2D Graph of Bonds
• 2.5D Surfaces
• 3D Coordinates
• 3.5D Conformers
• 4D Isomers, temporal evolution, etc

26
Chemical Informatics

• Informatics must be able to deal with
  variable-size structured data or convert data to
  standard vectorial format
• Graphical Models
• (Recursive) Neural Networks
• ILP
• GA
• SGs
• Kernels

27
Slide Title (Arial 44 pt)

• Font Arial 32 pt
• Font Arial 28 pt
• This Arial 24 pt
• 20pt
• Again 20 pt do not use font sizes lt 20 pt

Place useful information here i.e. Overview
28
SMILES Examples
29
SMILES Examples
30
SMILES Examples