Virtual screening and modelling: must it be atoms

1
Virtual screening and modelling must it be atoms?
Virtual screening and modelling must it be atoms?
Tim Clark Computer-Chemie-Centrum Universität
Erlangen-Nürnberg
2
What are molecules?
3
A Paradigm-shift?
4
Atoms in molecules (AIM)

• An approximation that relies on the
  transferability of properties of atoms and groups
  between molecules
• This requires transferability of the electron
  density assignable to an atom or group
• Follows from the first Hohenberg-Kohn theorem
• Made popular in the ab initio community by
  Richard Bader

5
AIM in virtual screening and modelling

• Fingerprints
• Fragment models
• Similarity (usually)
• Topological indices and descriptors
• Graph theory
• Atomic charge models
• Force fields
• Scoring functions
• Generalised Born solvation models

6
AIM in scoring functions
7
Scoring functions

• Desolvation free energies are probably at least
  as large as the complexation energy
• Two-center scoring increments assume transferable
  desolvation energies on both sides
• Is it any wonder we dont have a global scoring
  function?
• Why do we accept that scoring functions are local?

8
AIM in similarity searching

• Almost all classical methods are based on the
  bonding graph
• Carbo and Hodgkin indices are an exception
• They therefore find very similar bonding graphs
• May miss similar molecules
• Discriminate against scaffold hops

9
AIM in modelling force fields

• We usually use all-atom models
• United-atom force fields are limited and
  sacrifice accuracy
• All-atom models have two major disadvantages
• They scale badly
• They introduce high-frequency vibrational motion
  that doesnt interest us
• Short time steps
• Use SHAKE to remove (!)
• Vibrational partition function plays no role in
  the quantities that interest us

10
AIM in modellingelectrostatics

• Most force fields use point atomic multipoles
• Lead to two-center terms inseparable from
  dispersion/steric repulsion
• Overpolarise at short distances
• Are not properly shielded at long distances
• Must use fictitious and unphysical dielectric
  constants

11
Can we abandon AIM?

• Means moving to exclusively 3D methods
• No comfortable solution to the conformation
  problem

12
Can we abandon AIM?

• We need to know where the hydrogens are
• Which tautomer(s) are present in solution and
  bound to the receptor?
• Requires
• Systematic tautomer searching ?
• Very accurate pKa models ?

13
What do we need?

• Fast accurate generation of molecular surfaces
• Most consistent are isodensity surfaces
• These require the electron density (but not
  necessarily quantum mechanics)

14
What do we need?

• Ways to manipulate surfaces and surface
  properties quickly and efficiently
• Spherical harmonics
• Critical points
• Visual pattern recognition?
• PC-games technology (hardware and software)?

15
What do we need?

• Local properties to describe intermolecular
  interactions
• Molecular electrostatic potential for
  Coulomb-interactions
• Donor-acceptor?
• Dispersion?

16
What do we need?

• Intermolecular energy functions
• Surface-surface overlap
• Electrostatic, donor-acceptor, dispersion,
  repulsion
• If we include polarizability, these can be
  parameterised using ab initio data

17
What do we need?

• Anisotropic united-atom force field
• Monte-Carlo only needs energies
• Molecular dynamics needs
• MD in torsional coordinates
• Forces for anisotropic united atoms

18
What do we need?

• Surface-integral free energies
• Critical for scoring functions, which otherwise
  use the force-field intermolcular energies
• Provide an attractive alternative to
  descriptor-plus-interpolation QSPR-models
• Solvation ?, lattice energies ?, vapour pressures
  ?, partition coefficients ?, solubilities ?.....

19
Competence

• Aberdeen
• Spherical-harmonic surfaces, manipulation,
  superposition, docking
• Erlangen
• Quantum mechanics, local properties,
  surface-integral models, modelling
• Oxford
• Pattern-recognition
• Portsmouth
• Chemometrics, mapping, conformational searching
• Southampton
• Classical MD, sampling, pattern-recognition, free
  energies