High Performance Computing and Visualisation for Molecular Biology

1
Computational Vaccinology
Darren R Flower
http//www.jenner.ac.uk/res-bio.htm darren.flower
_at_jenner.ac.uk

2
COMPUTATIONAL VACCINOLOGY
Vaccines induce protective immunity, an enhanced
adaptive immune response to re-infection.
3
COMPUTATIONAL VACCINOLOGY
World class database Antigens, B cell and T
cell epitopes Peptide binding, Protein-Protein Int
eractions
Improved Prediction Class I and II T cell
epitope prediction B cell epitopes and Antigens
Experimental verification and data
discovery test prediction and generate new
binding data
4
T-cell, TCR and MHC
TCR, MHC and co-receptors on the surface of
T-cell and antigen-presenting cell. T-cells
have T cell receptors in their membranes that
bind to the protein fragments presented by the
MHC proteins. T cells recognise the presence of
foreign protein and hence pathogenic
micro-organisms and then destroy them.
5
TCR-peptide-MHC complex
Peptide MHC binding is just like the binding of
drugs to other receptors We can use QSAR and
molecular dynamics (MD) simulation to examine,
model and predict MHC-peptide interaction
6
DATA DRIVENMODELLING QSAR
Irini Doytchinova Channa Hattutawagama Valerie
Walshe PingPing Guan
7
QSAR
QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIP
STRUCTURAL DESCRIPTION and BIOLOGICAL RESPONSE
PREDICTIVE QSAR MODEL
ROBUSTMULTIVARIATE STATISTICS
8
Comparison of CoMFA and CoMSIA for HLA-A0201
r2pred lt 0.5 NC 6 q2 0.480 r2 0.911
r2pred 0.679 NC 5 q2 0.542 r2 0.870
9
Steric Map
Full CoMSIA Analysis of HLA-A0201
Hydrogen Bond Map
Electrostatic Map
Hydrophobic Map
NC 7 q2 0.683 r2 0.891 n 236
10
ADDITIVE METHOD FOR AFFINITY PREDICTION
HLA-A0201 NC 5 q2 0.337 r2 0.898 n
340
11
STRUCTURE DRIVEN MODELLING ATOMISTIC MOLECULAR
DYNAMIC SIMULATION
Shunzhou Wan
12
(No Transcript)
13
High Performance Computing Biomolecular
Simulations
Simulations of Biomolecular Systems
include proteins, nucleic acids, drug-receptor
interactions, protein folding, and a few
examples of more complex systems, such
protein-membrane interactions. Most simulations
done on desktop workstations and small
parallel machines (32 processors) Long time
scales and large systems generally
intractable HPC and the GRID allow us, for the
first time, to do things properly
14
Large Scale Molecular Dynamics

• Simulated systems are LARGE 30,000-300,000 atoms
• Simulation timescales are LONG In nanoseconds,
  even microsecond1
• Requires high performance computing
• We use scalable codes LAMMPS (Large-scale
  Atomic/Molecular Massively Parallel Simulator)
  NAMD
• on large parallel machines (up to 1000 nodes)

1. Duan Y, et al., Science 1998, 282740-744.
15
MD scaling performance (LAMMPS)
Parallelising the AMBER software scales very
poorly in our hands
16
MHC-peptide complexes
HLA-A0201MAGE-A4 complex Simulated using AMBER
force field in LAMMPS
17
MHC-peptide complexes What has been done?
a) ?1- ?2 domains periodic boundary no
constraints Rognan et al. (1992) Proteins 13,
70-85
b) c) ?1-
?2 domains periodic boundary constraints on
backbone Meng et al. (1997) Int. Immunol. 9,
1339-1346
all domains spherical boundary fix all atoms out
of sphere constraints on outer buffer region of
sphere Michielin et al. (2002) J. Mol. Biol.
324, 547-569
18
MHC-peptide complexes
Can the a3 and b2m domains and/or their
movement be neglected in simulations?
19
MHC-peptide complexes Simulation models

• Many authors1 regards this system as being out of
  reach of MD simulation
• "much too large"
• "relevant time scales inaccessible"
• But, with scalable codes and tightly coupled
  massively parallel machines ...

Partial model 30,574 atoms No constraints
Full model 58,825 atoms No constraints
Amber 98 Force Field
1. Nojima et al., Chem Pharm Bull (Tokyo) 2002
50(9), 1209-1214.
20
MHC-peptide complexes Simulation models
... for the 58,825 atom model (whole model), we
can perform 1 ns simulation in 17 hours' wall
clock time on 256 processors of Cray T3E using
LAMMPS
21
MHC-peptide complexes Results
For the partial system, about 300ps were required
for equilibration, while the whole system
required about 600ps, equilibration here being a
function of the size of the system.
RMS deviation from x-ray structure versus
simulation time (ps). Above partial MHC-peptide
system Below whole MHC-peptide system. Solid
line mainchain of protein Dotted line
mainchain of peptide.
22
MHC-peptide complexes Results
View of the b-sheets of the average structures
from the partial system simulation (blue) and the
whole system simulation (yellow), compared with
the x-ray structure (red). From top to bottom,
the sheets are juxtaposed from the N-terminal to
the C-terminal of the peptide. The view is
directly onto the peptide-binding side.
In the partial system simulation, the middle
sheets (?4, ?5) at the bottom of groove bulge
towards the peptide, while ?1, ?2 and ?8 turn
aside from it, whereas the whole system
simulation does not exhibit these effects.
23
MHC-peptide complexes Results
RMS deviations of peptide (top) and
antigen-binding site of MHC protein (bottom) from
x-ray structure. Solid line whole system
simulation Dashed line partial system
simulation.
The loop regions have large deviations from the
x-ray structure, while the two long helices and
the b-sheets have relatively small deviations.
24
Peptide Fluctuation vs Thermal B-Factor
B FACTOR
The larger deviations observed with the peptide
in the partial system indicate that the peptide
is considerably less tightly bound in the
partial system than in the whole system.
25
MHC-peptide complexes Conclusions

• For 58,825 atoms system, 1 ns simulation can be
  performed in 17 hours' wall clock time on 256
  processors of T3E.
• More accurate results are obtained by simulating
  the whole complex than just a part of it.
• The a3 and b2m domains have a significant
  influence on the structural and dynamical
  features of the complex, which is very important
  for determining the binding efficiencies of
  epitopes.

We are now doing TCR-peptide-MHC simulations (
100,000 atom model) using NAMD.
26
WE WANT USE MD TO ADDRESS FUNDAMENTAL PROBLEMS
IN IMMUNOLOGY
MHCs are polymorphic there are hundreds of
individual alleles with in the human population
each with a different peptide binding
specificity Use MD to identify binding epitopes
and use these to design and develop novel
vaccines Also want to use MD to examine more
complex systems such as the Immunological
Synapse which are not accessible to direct
experimental analysis
27
ACKNOWLEDGEMENTS
Bioinformatics Group Irini Doytchinova Shunzhou
Wan Helen McSparron Valerie Walshe PingPing
Guan Martin Blythe Debra Taylor
EJIVR Seph Borrow Outside Peter Coveney (UCL)
Funding from EPSRC (RealityGrid, CSAR) Jenner
Institute (GSK, BBSRC, MRC, DOH)