Transcript Structural Knowledge Base Development for Metal Complexes

Development of Molecular
Geometry Knowledge Bases from
the Cambridge Structural
Database
Stephanie Harris
Crystal Grid Workshop
Southampton, 17th September 2004
Cambridge Structural Database
 Stored geometric information for ~300,000 structures
 Search using Conquest
 Substructure search, user input required
Molecular Geometry Knowledge Bases
 Library of chemically well-defined geometric information
 Limited user input
 Rapid retrieval of statistical data
Molecular Geometry Knowledge Base:
 Mogul
 Bond lengths, valence angles and torsion angles
 Compiled from the CSD
Applications




Model building
Refinement restraints
Structure validation
Comparative values
Published bond length tables:




Organic and metal containing structures
Published late 1980s
Compiled from CSD of ~50,000 structures
Cannot be accessed by computer programs
Mogul 1.0




Whole molecule input
Graphical (cif, SHELX, mol2 files) or command-line interface
Integration with client applications, e.g. Crystals
Quick, automatic retrieval of statistical data, histogram
distributions, CSD structures
Search Algorithm
 All non-metal fragments in the CSD coded
 Set of keys code chemical environments
 Fragments with identical keys are chemically identical
 Use hierarchical search tree
 Generalised searching if insufficient hits
Mogul Search
.S1
.C7
Search
O
pTol
O
S
N
N
O
N
CN
Metal – Ligand Bond lengths
Me
C
O
Co-O bond length?
O
N
OH2
Co
N
OH2
O
C(O)Me
To be considered:
 Ligand type: Carboxylate
 Metal Oxidation State: Co(II)
 Metal coordination number: 6
 Ligand trans: Oxygen ligand
 Spin State?
Method
 Analysis of M-L bond lengths.
 For a range of metal and ligand types identify factors which
influence M-L bond lengths and evaluate their importance.
 For a defined Metal-Ligand group sub-divide bond
length distribution to produce ‘chemically meaningful’
datasets:
• Unimodal distributions.
• ‘Reasonably small’ sample standard deviations.
From hand-crafted examples develop an algorithm to produce a
molecular geometry knowledge base for metal complexes.
Data Tree
Metal-Ligand Group
Bin A1
Bin B1
Bin A2
Bin B2
Bin C1
Bin B3
Bin C2
Bin B4
Sharpened distributions
Smaller sample
standard deviations
Criteria Influencing M-L Bond Lengths
1.
Ligand, L
2.
Coordination mode of ligand
3.
Effective Metal Coordination Number
4.
Metal Oxidation State
5.
Metal clusters and cages
6.
Spin state
7.
Jahn-Teller effect
8.
Metal coordination geometry
9.
Ligand trans to L
M
=6
M
=6
Ligand Template Library
B
M A B
B
Ligand
• Non-metal atom or fragment bonded to a metal.
• Two ligands are the same if they have same connectivity
(topology) and stereochemistry.
OO- O
O
Method
• All ligands in CSD to be classified.
• Classify according to contact atom coordinated to metal.
• Ligands with multiple contact atoms can be present in
more than one ligand group. e.g. SCN-
Cambridge Structural Database
 Approximately 22,000 formulae
 Approximately 780,000 ligands
No. of occurrences of
unique formulae in CSD
Total Number of
Ligands
Number of formulae

550,000 (70%)
70
100 – 999
109,263 (14%)
394
10 – 99
76,000 (10%)
3000
1–9
45,700 (6%)
18,937
Ligand Template Hierarchy
• Exact ligand templates (724)
• R-substituted templates (H’s replaced with ‘innocent’ R groups)
• Generic templates (ALL ligands classified)
Cobalt Carboxylate Bond Lengths
Co
O
3
C C sp
No. of
Frags.
O
Co-O: 1.929(62) Å
619 Fragments
Co-O (Å)
Co
O
3
C C sp
O
Co(II)
Co(III)
2.049(58) Å
1.904(20) Å
1.929(62) Å
OC(O)C
L
L
Co II
L
L
L
2.073(42) Å
1.904(20) Å
OC(O)C
L
L
Co III
L
L
L
1.910(15) Å
OC(O)C
L
L
Co II
L
L
O
2.074(32) Å
OC(O)C
L
L
Co III
L
L
N
OC(O)C
L
L
Co III
L
L
O
1.895(17) Å
Fe-Cl
 Chlorides
2.242(68) Å
Cl
III
Fe L
L
L
2.189(24) Å
 Pyridines e.g. Fe
(spin state)
Fe N
Fe(II)L5py
High Spin
2.166(84) Å
2.225(29) Å
 Tertiary phosphines, Carbon-ligands
 Copper complexes (Jahn-Teller effect)
Standardisation of Cu connectivity
Cu(II)-OH2
2.232(225) Å
Metal-Ligand Knowledge Base
1. CSD data adjustment:
 Standardisation of metal connections
 Assignment of metal as part of a metal cluster
 Assignment of metal oxidation state
2. Classification of ligands by ligand template library
3. Perform algorithm on all possible M-L fragments to produce
knowledge base
Algorithm:
Metal-Ligand Group
From ligand template library:
Generic or more specific
e.g. Carboxylates:
C
O
O
O
O
C
C
O
3
sp
C
C
O
Et
Metal-Ligand Group
‘Metal Clusters’
Division on Oxidation State
Division on Metal effective coordination number
Division on spin and Jahn-Teller effect
• Only for particular metals, oxidation
states and coordination numbers.
• Not found for all ligand types.
• Not searchable in CSD.
Flag users, effects evident by:
bimodal histogram, high SSD, outliers.
Metal-Ligand Group
‘Metal Clusters’
Division on Oxidation State
Division on Metal effective coordination number
Division on spin and Jahn-Teller effect
Division on Metal coordination geometry
E.g. 4-coordinate geometry:
Tetrahedral, square planar, disphenoidal
Metal-Ligand Group
‘Metal Clusters’
Division on Oxidation State
Division on Metal effective coordination number
Division on spin and Jahn-Teller effect
Division on Metal coordination geometry
Divide on trans ligand to L
More specific ligand
e.g. alkyl carboxylate
Final Ligand division
Generalised Searching
• No hits or insufficient number of hits.
• Allows the retrieval of data on related fragments.
• Hierarchical search tree structure
• Move up to a higher, less specific level of data tree.
• Order of algorithm important.
 Should order of criteria be changed?
 Should order depend on M-L group?
E.g. Should oxidation state always be the first main
division?
Conclusions
• Pre-processing of structural data from the CSD to construct
molecular geometry knowledge bases.
• Knowledge bases to contain chemically well-defined datasets.
• Limited user input required.
• Quick, automatic retrieval of statistical data, distributions.
• Efficient analysis of large number of chemical fragments.
• Outliers, high SSD?
 Further Analysis – Computational Chemistry.
• Further development to include extra chemical information
e.g. computational data.
Acknowledgements
Bristol University:
Guy Orpen
Natalie Fey
X-Ray Crystallography Group
Cambridge Crystallographic Data Centre:
Robin Taylor
Frank Allen
Ian Bruno
Greg Shields