Generating Synthetically Accessible Ligands by De Novo Design

1
Synthetic Sprout

• Generating Synthetically Accessible Ligands by De
  Novo Design

A Peter Johnson Krisztina Boda Attilla Ting Jon
Baber
2
SPROUT is the De Novo design system developed in
Leeds

• SPROUT components
• Identification of potential interaction sites
  complementary to the receptor, ie H bonding,
  hydrophobic sites, metal co-ordination sites etc.
• Automated docking of small fragments at the
  interaction sites.
• Generation of hypothetical structures by linking
  the docked fragments together.
• Tools for scoring, sorting and navigating the
  answer set.

3
Hydrogen Bond Sites
H-bond acceptor site
H-bond donor site
Example 3D shapes of sites
4
(No Transcript)
5
Boundary Surface
6
Docking of small fragments at target sites

• Target sites are generated either by SPROUT
  module HIPPO (or similar system) or come from a
  pharmacophore hypothesis.
• Small fragments with complementary functionality
  are selected by the user and automatically docked
  into the target site(s).
• In addition to these small fragments, it is also
  possible to dock large fragments which are known
  to satisfy several of the target sites. Such a
  large fragment can then act as a seed for
  further growth.
• A successful dock must place the small fragment
  at the target site with the correct orientation
  to satisfy any directional constraints.
• The docking process is very fast and uses a novel
  hierarchical least squares optimisation
  procedure.

7
Structure generation

• The SPIDER module links the target sites together
  in a pairwise fashion to make complete molecular
  structures which satisfy target sites. It does
  this by sequentially adding new fragments in an
  exhaustive fashion.
• There is no element of random choice in this
  process, which means that various heuristics have
  to be adopted to avoid a combinatorial explosion.
• The main approximations employed are
• There is a sampling of all the possible
  conformations about single bonds.
• Growth is only permitted from atoms/bonds which
  are closest to the target site which is to be
  reached

8
Main algorithm of SPIDER

• Multiphase heuristic graph search on a forest (
  set of trees)
• Two trees are searched and removed in each phase
  and a new tree generated which contains skeletons
  connections both set of sites
• Each phase consists of
• a bi-directional search
• Breadth First Search (BFS)
• Depth First Search (DFS)

Typical saving bi-directional search 10
successors, 6 level 2x103 ltlt 106
9
Connection of Partial Structures

• Common template is located in two structures
  (one from each tree)
• Structures are overlayed by the common template
• Combined structure is docked to the united set
  of target sites also considering the steric
  constraints of the receptor site
• Side effect joins are axamined for validity
  (e.g. fusion on figure)

10
Navigating the answer sets

• Estimated binding energy score
• Ranking final de novo set
• Ranking and pruning (with caution) intermediate
  trees to reduce combinatorial problem.
• Estimated ease of synthesis score
• Ranking final de novo answer set
• Too slow (1 structure per minute) to be useful
  for intermediate pruning
• Need faster methods for intermediate pruning

11
Recent Advances

• Parallelization of structure generation
• Farm of SGs or pcs
• SPROUT server BEOWOLF cluster currently 11 dual
  processor 600Mhz Pentium III
• VLSPROUT screens virtual libraries
• SYNSPROUT generates synthetically accessible
  ligands
• Receptor SPROUT generates potential synthetic
  receptors for small movecules

12
The perennial modellers problem

• Hypothetical ligands, including those predicted
  to bind very strongly, have no practical value
  unless they can be readily synthesised.
• Our attempts to provide solutions
• CAESA post design estimation of synthetic
  accessibility
• SynSPROUT synthetic constraints built into the de
  novo design process
• VLSPROUT even greater synthetic constraints
  only members of a specific virtual library are
  generated

13
Synthetic Sprout Approach
Pool of readily available starting materials,
e.g. subset of ACD
Knowledge Base of reliable high yielding
reactions, e.g. esterification, amide formation,
reductive amination..
VIRTUAL SYNTHESIS IN RECEPTOR CAVITY
Readily synthesable Putative ligand structures
14
Creation of Starting Material Libraries

• Obvious Classes eg amino acids
• Drug like starting materials selected by hand
• Drug like starting materials generated
  automatically by retrosynthetic analysis of drug
  databases

15
Retro-Synthetic Knowledge BaseRetro-Synthetic
Rule
16
Automatic Template LibraryGeneration

• Perception
• Knowledge
• Bases
• Aromatic
• Normalisation
• Hybridisation
• H-bonding
• properties

2D Drug-like Structures
Ring Perception
Fragmentation
Retro-Synthetic Knowledge Base
Clustering
Retro-synthetic patterns
Filter
Retro-Synthetic rules
Single 3D Conformer Generation
Corina

• Synthetic
• Knowledge Base
• Functional groups

Multiple Conformer Generation
Omega
Synthetic Template Library
17
Automatic Chemical Perception
Rule based system where rules are encoded using
the PATRAN language (similar to SMILES)

• Information Perceived
• Aromatic atoms and bonds
• Normalised bonds
• Hybridisation including induced hybridisation
• H-Donors / Acceptors
• Number of hydrogens attached to an atom
• Number of connections to an atom
• Number of available electron pairs
• Charge at an atom

Example from Hybridisation knowledge base
CHEMICAL-LABEL ltNitrogenWithLP--SP2gt XSPCENTRE2
-NHS0,1,2SPCENTRE3 EXPLANATION N with lone
pair next to sp2 centre behaves as sp2. IF
NitrogenWithLP--SP2 THEN set-av-eps 2 to 0
set-hybridisation 2 to 2 END-THEN
18
Perception - Binding Properties

• O Single atom based
• Vs
• C Functional group based
• D - H donor
• A - H acceptor
• J - Joinable
• H - Hydrophobic
• N - None

O - original method C - current method
According to reaction knowledge base
19
Synthetic Template
Primary Amine (Donor)
D
A
A
A
H
AD
A
H
A
Carboxylic Acid (Acceptor)
Phenol (Acceptor-Donor)
20
Synthetic Knowledge BaseSynthetic Rules
EXPLANATION Amide Formation 1 IF Carboxylic Acid
INTER Primary Amine THEN destroy-atom 3
form-bond - between 1 and 5
change-hybridization 5 to SP2 Dihedral 0
0 Dihedral 0 180 Bond-length 1.35 END-THEN

• Joining Rules
• Steps of formation
• Hybridization change
• Bond type
• Bond length
• Dihedral angles/penalties

21
De-novo DesignUsing Synthetic Sprout
Donor site
Acceptor Site
1.Amide Formation ( Carboxylic Acid -Primary
Amine )
2.Reductive Amination ( Carbonyl - Primary Amine )
22
New Problems - Hybridisation change (SP3? SP2)
Secondary Amine Nitrogen becomes SP2
Hybridisation change in Amide Formation 2. (
Carboxylic Acid - Secondary Amine )
23
Hybridisation change (SP2? SP3)
Carbonyl Carbon becomes SP2
Hybridisation change in Reductive Amination 1. (
Carbonyl - Primary Amine )
24
Selection of Synthetic Reactions

• Amide Formation
• Ether Formation
• Ullman reaction
• Amine Alkylation
• Ester Formation
• Aldol
• Wittig
• Imine
• C-S-C Formation
• Reductive Amination

25
CDK2
Library 300 fragments/1055 conformations
Run time 10 h
26
SynSPROUT

• Current status
• Works well for small starting material libraries
  (low hundreds).
• Several libraries now built including amino acid
  library for peptide generation. Library from MDDR
  being built.
• Potential for suggesting starting points for new
  combinatorial libraries
• Future work
• Extend types of chemistry allowed
• Develop algorithms which would permit the use of
  libraries of hundreds of thousands of starting
  materials (such as ACD).
• Parallelisation helps but on its own is not
  sufficient to cope with the inevitable
  combinatorial explosion.

27
Acknowledgements
Co-workers Krisztina Boda Attilla Ting Jon
Baber Special thanks to Open Eye Scientific
Software for providing access to OMEGA