GBSA And Other Bad Solvation Models

1
GBSA-And Other Bad Solvation Models

• Anthony Nicholls
• CUP IV, 2003

2
Basis of Generalized Born Theory
Solvent Energy Missed
Dipole Size (E Q/r2) Back Potential
(1/r2) ? Energy Deficit ? (1/r4)
3
Generalized Born Equation
Self-Energy (kcals) -166.0q2
(1.0-(1/esolvent)) (1.0/radius) - ltmissing
solvent/r4gt
-166.0q2 (1.0-(1/esolvent)) /a
Interaction Energy 166.0q1q2 1.0/ri
j - (1.0-(1/esolvent)) /sqrt(rij2
a1a2exp(-a1a2/(4rij2))
4
Neutral CSD Subset Modern GB
5
(No Transcript)
6
(No Transcript)
7
Original GB Test Set Revisited
8
From Integrals to Sums
9
Error Analysis of MacroModel GB
10
Error Analysis GB vs PB-VdW surface
11
Error Analysis Hard Sphere GB
12
(No Transcript)
13
1.0 Angstrom Spacing
Illustration of PB Grid Scale
1.6 Angstrom Spacing
14
Gaussian Generalized Born
Suggests
OR
15
Error Analysis Gaussian GB
16
The Effect of Correct Born-Radii
17
Why Dont Exact GB Radii Help?
1) Self and Pair Terms Errors Compensate 2) Self
and Pair Terms are Both Large 3) Pair term still
has poor directionality
18
Alternate Pair Term
19
Error Analysis Alternate Pair Term
20
Algorithmic Complexity
1) GB and Molecular Mechanics, are NAIVELY O(n2).
GB doubly so.
2) MM at least has rigorous error analysis on
CUT_OFFS. GB has none.
3)PB (via MultiGrid) is O(n)
21
Electrostatic Separation (Gaussian)
22
You Make The Call - I

• GBSA is heavily parameterized, PBSA largely not
• Even then, GBSA has large outliers
• Does have 2nd-order derivatives
• GBSA faster (just) for small molecules, slower
  for large
• Wildly inaccurate for proteins

23
Sheffield, Home to

• World Snooker Championships
• The Full Monty
• Some Excellent Chip Shops
• A Bad Solvation Model

24
The Sheffield Solvation Model

• Brought to you by
• Barry Pickup, Matthew Sykes,
• Christine Kitchen and Andrew Grant

25
Eighteen Blockbuster Drugs
26
Test System Cyclic Urea
27
Sheffield vs ZAP Cyclic Urea.Multiple
Conformers (50), Single Charge Set (MMFF).
28
Sheffield vs ZAP Cyclic Urea.Multiple Charge
Sets (5168), Single Conformer.
29
Sheffield vs ZAP Cyclic Urea.Multiple Charge
Sets (5168), Single Conformer, Close up.
30
ZAP vs Sheffield (Tweedledum)ZAP vs GBSA
(Tweedledee)
Merck Test Set lt 100 kcals
31
Just How Similar are They?
32
What About Proteins?
33
Protein Solvation Energy
34
What about Binding to Proteins?
35
You Make the Call - II
for(i0iltAtomNumi)
for(j0jltAtomNumj)
dist(coords3i -coords3j )(coords3i
-coords3j )
(coords3i1-coords3j1)(coords3i1-coor
ds3j1) (coords3i2-coords
3j2)(coords3i2-coords3j2)
Solvation -chargesjchargesi/sqrt(AdistB
radiijradiii)
VS.

• Thousands of lines of code
• Hundreds of papers
• Score of wasted graduate students
• Columbia owning your implementation

36
What about Experiment?

• Data Set of 200 polar molecules
• (Courtesy of Rob Rizzo, UCSF)
• ZAP Using GAST MMFF AM1BCC AM1BCC_U
• RMS Expt (kcal) 3.62 2.68 1.55
  1.31
• With AM1BCC_U
• ZAP 1.31
• GB 1.491
• Sheffield 1.495

37
Can GB ever, in any conceivable, possible
Universe be a better theory than PB?

• Is it a better theory if it gets a better
  correlation coefficient to experiment?
• AM1, Liquid Theory,Mean-field Theories, PB(!) and
  parameters.
• Adding parameters may make a more useful
  theory, but not a better theory
• A real theory predicts AND explains.

38
Conclusions

• STOP
• THE MADNESS!

39
Deep Appreciations to..

• The Sheffield Gang- Andrew Grant, Christine
  Kitchen, Matthew Sykes, Barry Pickup
• Rob Rizzo, UCSF
• Kim Sharp, Paul Labute, Barry Honig