OEChem in PubChem: Cleaning Up PDB Small Molecules

1
OEChem in PubChem Cleaning Up PDB Small
Molecules

• Paul Thiessen
• NCBI

2
Outline

• Introduction
• Legacy data
• Small molecules from PDB
• OEChem perception from 3D
• Successes
• Successes, with some tweaking
• Outright failures
• Molecule classification
• Bound vs. free ligands
• Blacklist
• Other projects

3
Crosslinks

• Specialized legacy data sets
• Open BioCyc
• KEGG
• PDB (MMDB)
• lt5 of structures in PubChem, but 100 of
• Direct links from chemicals to
• Literature (PubMed, except MeSH)
• Sequence (GenBank/GenPept)
• Protein structure (MMDB)
• Enzyme activity (EC)

4
Sample

• ATP from KEGG

5
PDB Molecules

• What is a small molecule in PDB?
• Any non-biopolymer
• A protein of lt 10 residues
• A nucleotide of lt 5 residues
• Dont distinguish bound vs. co-crystallized
• Whatever PDB puts in a HET residue
• Hydrolyze covalent bonds (no R-groups)
• Ignore uninteresting molecules
• Solvents
• Monoatomics
• Blacklisted

6
The Problem

• PDB lacks chemical detail
• No bond orders
• No charges
• Mostly no hydrogens

7
Filling in Detail
?

• So how do we go from this

to this?
8
OEChem

• The answer OEChem
• OEChem 1.3 by OpenEye
• Chemical information
• Manipulation of chemical structure
• Perception from 3D coordinates

9
Perception Results

• OEChem sets bond orders, hydrogens, and charges

to the deposited structure
Now we can go from PDB
10
Standardization

• The next step (also with OEChems help)

Deposited structure has bond orders and
(implicit) hydrogens
Standardized structure includes stereochemistry
(perceived from 3D coordinates)
11
Success

• Some complex molecules come out right, just on 3D
  coordinates alone
• Vancomycin

12
Failure

• Some dont (noise in PDB)

13
Spatial Overlap

• Sometimes PDB overlaps molecules

14
Success (Mostly)

• Some need a little massaging

Pattern match to fix heme bonds, oxidation state
15
Hard Cases

• Without H, bond order, and charge ?

16
Diatomics

• Diatomics are also hard to distinguish
• Especially when bound to a metal (heme)
• OO or HO-OH?
• Similarly for
• Cyanide (vs. aminomethane)
• Carbon monoxide (vs. methanol)
• Nitrogen (vs. hydrazine)
• Nitrogen monoxide (vs. hydroxylamine)
• Other special functional groups
• Isocyanides (M N-C-R)
• Azide (N-NN-)

17
Carbocations

• Sometimes we get charged carbons

18
Carbanions

• OEChem wont mix implicit/explicit H

19
Resonance

• Sometimes OEChem chooses an unlikely resonance
  form

20
Errors?

• Not always clear why OEChem does what it does

21
Bad Kekulization

• Some cases fail to Kekulize properly

22
OEChems Valence Model

• Phosphorus with valence 4 can be /- 1

23
Valence Model (contd)

• Oxygen 1 is allowed valence 3, 5

24
PDBs HET

• PDB sometimes splits into separate HETs

Bonds get hydrolyzed
25
PDBs HET (contd)

• This can be a bit arbitrary

26
Statistics

• Latest PDB data set
• 22,075 substances
• 6,791 unique substances (by connectivity)
• Ignored 7,042 blacklisted molecules
• 3,075 sulfate ions
• 979 glycerols
• 882 phosphate ions
• 567 acetate ions
• 323 mercaptoethanols
• 314 ethylene glycols
• PubChem data is freely available!

27
PCView
28
Thanks To

• Steve Bryant
• PubChem Team
• NCBI SWG
• OpenEye
• CUP-VI Organizers