Dealing with organometallic molecules in RDKit

My talk the at 2020 RDKit user group meeting 2020.10.06

Transcript

1. Dealing with organometallic molecules in RDKit Jan H. Jensen Department of Chemistry, University of Copenhagen @janhjensen 1 2020 RDKit UGM 2020.10.06
2. Most homogeneous catalysts are organometallic compounds Large datasets are becoming available but in xyz format Most cheminformatics/ML relies on SMILES/graphs (e.g. substructure searching and graph convolution) chemrxiv.12894818.v1
3. J Cheminform (2018) Fe H2 O H2 O HO OH2 OH H2 O [OH2][Fe](O)(O)([OH2])([OH2]) [OH2] Not readable by RDKIT Charge from SMILES often incorrect
4. xyz2mol for organic compounds xyz2mol N N N Cl Cl O O O Si N H O N N N Cl Cl O O O Si N H O xyz2mol converts an xyz file to an RDKit mol object (needs the molecular charge and hydrogens) github.com/jensengroup/xyz2mol
5. Organic examples X H H H H H H H H * * [4, 4, 1, 1] H H H H valence N H H H O O N H H H O O N H H H O- O- [2, 4, 1, …] N H H H O O * * N H H H O O N+ H H H O O- [1, 3, 1, …]
6. Sometimes there are more than one solution xyz2mol will generate one of them arbitrarily Solution(?): generate all, then filter Generate all using rdchem.ResonanceMolSupplier* Create filter that picks “canonical” form *HT Mads Koerstz
7. One approach for organometallics Distinguishing dative from covalent bonds Fe H2 O H2 O HO OH2 OH H2 O Fe OH H2 O HO OH2 OH H2 O Fe+2 Fe+3 Formal charge on Fe = total charge [e.g. Fe(OH)2 +] O->[Fe](O)(O)(<-O)(<-O)<-O O->[Fe](O)(O)(O)(<-O)<-O
8. Main problem Distinguishing dative from covalent bonds before bond orders are assigned Fe N P H2 N P H2 C O OH Me Fe N P H2 N P H2 C O O Me DFT: 1.97Å DFT: 1.94Å
9. Another approach Only dative bonds Fe+2 Fe+3 Formal charge in Fe = total charge + ∑ charge on ligands Also not sensitive to presence of bond(s) Fe+2 H2 O H2 O -HO OH2 OH- H2 O Fe+3 OH- H2 O -HO OH2 OH- H2 O Alex Clark J. Chem. Inf. Modeling 2011
10. Must know charge on Fe Fe N- P H2 N- P H2 C O OH Me Fe N- P H2 N P H2 C O O- Me total charge = charge on Fe + ∑ charge on ligands
11. Most TMs have many different oxidation states https://byjus.com/chemistry/transition-elements-oxidation-states/
12. Try all charges and save cases for which total charge = charge on TM + ∑ charge on ligands W+4 N- C H - C- O C- W+6 N-2 C H - C- O- C- W+4 N- CH- C- O -H2 C Paper Expect (W+4) Get W+2 N C H - C- O+ C- W+4 W+2 W+6 Missing bond to W “wrong” resonance form of ligand
13. Some issues Not all bonds to TM are found (uses RDKit Hückel reduced overlap population) Other resonance forms of ligands Hydrides
14. Dative bond limits ResonanceMolSupplier Fragment -> Resonance structures -> combine(?)
15. Hydrides and SMILES (RDKit 2020.03.5)
16. Hydrides and SMILES (RDKit 2020.03.5) SMILES does not give mol object with correct charge [HH2-]->[Fe+2] [FeH+2]or [H][Fe+2] becomes
17. Hydrides and SMILES (RDKit 2020.03.5) Greg found a workaround Another option is to treat TM-hydride bonds as covalent and reduce TM charge [FeH+1] [HH2-]->[Fe+2] instead of
18. Summary Prototype generates RDKit readable SMILES for organometallic compounds w/o human intervention But … Not all bonds to TM are found (uses RDKit Hückel reduced overlap population) Non-unique oxidation states/resonance forms (filter/”canonicalization”?) Hydrides charge bug for MolFromSmiles How to automatically test code?
19. Fe N- P H2 N- P H2 C O OH Me Additional RDKit issues Depiction of octahedral compounds not helpful Embedding/UFF optimization not working
20. https://github.com/jensengroup/xyz2mol/tree/tm_comb Experimental branch can be found here Feedback welcome Additional issues continued Specifying stereochemistry https://www.simolecule.com /cdkdepict/depict.html