AUSTRIAN GRID 1/14 AUSTRIAN GRID TECHNICAL REPORT 1 Document Identifier: AG-DA-1b-1-2005_v1.doc Date: 2005-07-18 Workpackage: A-1b WP Leader: Schreiner
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
AUSTRIAN GRID
1/14
AUSTRIAN GRID
TECHNICAL REPORT 1
Document Identifier: AG-DA-1b-1-2005_v1.doc
Date: 2005-07-18
Workpackage: A-1b
WP Leader: Schreiner
AUSTRIAN GRID
2/14
Delivery Slip
Name Partner Date Signature
From
Verified by
Approved by
Document Log
Version Date Summary of changes Author
0.1 2005-07-04 Milestone “Software Selection” U. Omasits
0.2 2005-07-05 other Milestones U. Omasits
0.3 2005-07-06 Figures U. Omasits
1.0 2005-07-18 adapt report to template U. Omasits
AUSTRIAN GRID
3/14
1 ABSTRACT ................................................................................................................................................. 4
2 MILESTONE – SOFTWARE SELECTION............................................................................................ 5
2.1 BENCHMARK SYSTEMS AND RESULTS .................................................................................................. 5 2.2 INSTALLATION NOTES AND DECISION................................................................................................... 9 2.3 PARALLELIZATION OF GROMACS .......................................................................................................... 9
3 MILESTONE – SELECTION OF PMHC-COMPLEX ........................................................................ 12
4 MILESTONE – SELECT GRAPHICS SOFTWARE............................................................................ 14
5 MILESTONE – PRODUCE GRAPHICS ............................................................................................... 14
AUSTRIAN GRID
4/14
1 Abstract This document lists the proceedings according to the milestones of the pMHC-complex
molecular dynamics project. The software has been evaluated and Gromacs was chosen for
the pMHC simulations. The parallelization of Gromacs was also tested. A list of simulateable
pMHC complexes was generated and binding-affinity data was collected. The pMHC-
complex with which the first simulations shall be performed was selected. Graphic software
was tested and a short animation was produced.
AUSTRIAN GRID
5/14
2 Milestone – Software Selection
Milestone Date Description
Focus 1a:
Software Selection 2005-03-31
Several Packages for Molecular Dynamics simulation shall be evaluated and a decision be made.
Evaluate which type of parallel equipment is available among GRID-partners.
Use Test-System (Waterbox 8.5k + ubiquitin)
Check CHARMM, NAMD & Gromacs for performance on ANAX (4 Xeon shared memory) and Beowulf & & SGI.
If possible regarding performance choose → Charmm as the most versatile package.
Professor M.Neumann (Dept. of Experimental Physics) and Prof. O.Steinhauser (Dept. of
Structural BioChemistry, both University of Vienna) and DI Rene Kobler (University Linz)
have installed and evaluated the three, above mentioned packages.
2.1 Benchmark Systems and Results
NAMD (University of Illinois at Urbana-Champain) and CHARMM (Harvard University)
were tested with the JAC1000 benchmark system (Joint Amber-Charmm) instead of the
ubiquitin box because it is matching with our target more closely – a pMHC in solvent.
Unfortunately the JAC benchmark is not available for GROMACS (University of Groningen),
so in Vienna the PDB-File 1ROG was used instead, consisting of a similar number of total
atoms. In Linz the DPPC-benchmark was used with GROMACS. Additionally, NAMD was
tested with the ApoA-I-benchmark. The results are summarized in the table(s) below.
JAC1000 – protein in water; 23558 = 21069 water + 2589 atoms, 1000 timesteps
1ROG.pdb – pMHC complex; 25623 atoms, 5000 timesteps, cutoff
DPPC – a membrane system; 121856 atoms
ApoA-I – a large system with good scaling capabilities; 92442 atoms
AUSTRIAN GRID
6/14
Result of
Benchmarks to select one MD-package for
simulating the interaction between epitope-MHC and pMHC-TCR
Results in seconds and speed-up in brackets for test-systems
Hardware: Vienna – ANAX (4-prozessor Intel shared memory)
Tests performed by: Professor M.Neumann
Note: All the execution times are arithmetic means of ten runs.
criterion CHARMM (charm-3.1b1 + g77 + mpich)
Gromacs (gromacs-3.2.1 + g77 + lam)
NAMD (namd-2.5/charm++)
test-system JAC1000 1ROG.pdb JAC1000
np=1 2590 1225 1807
np=2 1443 (1,80) - 1404 (1,29)
np=4 1108 (2,34) 360 (3,40) 1097 (1,65)
Hardware: Linz – SGI Altix 350 (64 processor cluster, 4 nodes, each of them consisting of 16 processors)
Tests performed by: DI Rene Kobler
Note: All the execution times are arithmetic means of eight runs.
criterion CHARMM c3b1 Gromacs 3.2.1 NAMD (namd-2.5/charm++)
NAMD (namd-2.5/charm++)
test-system JAC1000 DPPC JAC1000 ApoA-I
np=1 566,13 8041,08 610,8 2169,1
np=2 355,88 (1,59) 5499,63 (1,46) 328,90 (1,86) 1110,40 (1,95)
np=4 170,63 (3,32) 3064,66 (2,62) 164,90 (3,70) 565,10 (3,84)
np=8 110,50 (5,12) 1799,13 (4,47) 92,20 (6,62) 303,65 (7,14)
np=16 81,50 (6,94) 1021,88 (7,87) 49,63 (12,31) 169,40 (12,80)
np=32 - 1053,75 (7,63) 472,90 (1,29) 146,60 (14,80)
np=48 - - - 123,80 (17,52)
np=64 - - - 147,60 (14,70)
Hardware: Linz – Hydra Cluster (16 processor Athlon MP cluster, 8 Dual-Athlon MP nodes)
Tests performed by: DI Rene Kobler
Note: All the execution times are arithmetic means of four runs.
criterion Gromacs 3.2.1 NAMD (namd-2.5/charm++)
test-system DPPC JAC1000
np=1 9371,80 832,45
np=2 5549,80 (1,69) 466,95 (1,78)
np=4 3663,30 (2,56) 273,81 (3,04)
np=8 2275,30 (4,12) 160,89 (5,17)
np=16 1430,80 (6,55) 121,59 (6,85)
AUSTRIAN GRID
7/14
Following figures show the speed-up of the different molecular dynamic packages with the
various benchmark-systems:
SGI Altix 350 - speed-up
processors
2 4 8 16 32 48 64
speed-up
5
10
15
20
5
10
15
20NAMD - ApoA-INAMD - JAC1000Gromacs - DPPCCHARMM - JAC1000
Figure 1: Speed-up of NAMD (two benchmarks), Gromacs and CHARMM (each one benchmark) on the
SGI Altix 350
AUSTRIAN GRID
8/14
Hydra Cluster - speed-up
processors
1 2 4 8 16
speed-up
2
4
6
8
2
4
6
8NAMD - JAC1000Gromacs - DPPC
Figure 2: Speed-up of NAMD and Gromacs on the Hydra Cluster
Benchmark-Results of the following webpage were also discussed and were taken into
consideration. http://amber.scripps.edu/amber8.bench2.html
Speed-up of the JAC1000-benchmark according to
http://amber.scripps.edu/amber8.bench2.html
processors CHARMM c31a2 Amber 8 NAMD 2.5
2 1,7 1,9 1,8
4 3,4 3,7 3,4
8 5,6 7,1 6,0
16 8,9 13,3 10,3
32 11,2 23,4 17,2
64 10,8 36,3 24,8
This table shows that speed-up can be maximized but with a lot of configuration- and
optimization-expenditure – and this is not the intention of this project.
AUSTRIAN GRID
9/14
2.2 Installation Notes and Decision
NAMD was difficult to install and had serious problems with parallelization. On the SGI
Altix NAMD was only parallel runable after script modifications. On the Hydra-Cluster only
a pre-built version worked.
CHARMM was easy to install on both machines (SGI Altix 350 and Hydra Cluster) but was
only executable on the SGI Altix 350. In Vienna (Anax) CHARMM will not work with
LAMMPI but with MPICH.
GROMACS was easy to install too. There were no problems, neither in Vienna nor in Linz.
Gromacs has a special and very fast procedure to treat water molecules. Because we will need
large water-shells for our simulations and because of the good documentation Gromacs will
be used.
2.3 Parallelization of Gromacs
Professor M.Neumann studied the parallelization behaviour of Gromacs using two different
methods to simulate electrostatics and VdW-interactions (cutoff vs. PME). Additionally he
varied the thickness of the water shell.
Result of
Benchmarks to examine the parallelization of Gromacs Hardware: Linz – Altix2 (single precision)
Tests performed by: Professor M.Neumann
test-system: 1ROG.pdb
Note: All the values are single-measured execution times. There are a few faulty values due to other running processes on the machine, but without these values (marked with * ) a trend is clearly visible in the figures below.
cutoff (see figure 3)
water-shell thickness and resulting atom-numbers
d=10A d=15A d=20A d=30A
processors n=25633 n=39916 n=59548 n=111277
1 924,000 1.297,000 1.760,000 2.957,000
2 612,000 819,000 1.064,000 1.648,000
3 519,000 666,000 848,000 1.247,000
4 473,000 583,000 728,000 1.037,000
5 457,000 537,000 651,000 889,000
6 424,000 508,000 597,000 813,000
7 410,000 1.523,000* 558,000 730,000
8 402,000 1.997,000* 556,000 697,000
9 388,000 547,000* 529,000 684,000
10 385,000 453,000 705,000* 630,000
11 396,000 449,000 1.112,000* 636,000
12 404,000 430,000 490,000 604,000
13 398,000 420,000 484,000 601,000
14 393,000 415,000 464,000 561,000
15 398,000 423,000 458,000 593,000
16 394,000 424,000 466,000 564,000
AUSTRIAN GRID
10/14
PME (see figure 4)
water-shell thickness and resulting atom-numbers
d=10A d=15A d=20A d=30A
processors n=25633 n=39916 n=59548 n=111277
1 1507,00 2726,00 3580,00 6052,00
2 943,51 1282,33 1886,77 3126,00
3 738,02 1006,05 1667,16 2215,00
4 668,44 865,77 1357,08 1819,00
5 613,31 821,39 1008,01 1633,00
6 582,27 730,82 931,29 1580,00
7 547,05 690,66 854,81 1390,00
8 539,53 676,84 828,69 1350,00
9 726,94* 690,33 838,67 4411,00*
10 1464,69* 645,92 811,15 1195,00
11 514,12 628,85 804,55 1113,00
12 614,94 619,84 756,07 1100,00
13 532,27 625,41 745,89 1082,00
14 534,57 587,14 710,86 1074,00
15 546,10 623,03 746,99 1041,00
16 550,90 637,05 726,24 1059,00
* values were left out in figures
simulation speed-up - with cutoff
processors
2 4 6 8 10 12 14 16
speed-up
1
2
3
4
5
6
d=30Åd=20Åd=15Åd=10Å
Figure 3: Speed-up of Gromacs using “cutoff” with four different water-shell thicknesses on the Anax
AUSTRIAN GRID
11/14
simulation speed-up - with PME
processors
2 4 6 8 10 12 14 16
speed-up
1
2
3
4
5
6
d=30Åd=20Åd=15Åd=10Å
Figure 4: Speed-up of Gromacs using “PME” with four different water-shell thicknesses on the Anax
As can be seen from figure 3 and figure 4, PME in general allows a far higher speed-up.
Although PME is slower (about 10 to 50%) than cutoff, we will use PME for further
simulations because it is more accurate and better scaleable.
The bigger the water-shell the greater the speed-up. This is due to the fact that larger problem
sizes in general yield better scaling capabilities and maybe also due to the special procedures
for water in Gromacs.
All the benchmark data show that – in order to optimize the computational time – we will run
two simulations parallel on one of the 16 processor machines (8 processors each job) rather
than only one simulation because the speed-up always flattens between 8 and 10 processors
and even decreases in most cases above 16 processors because of the inter-node
communication.
AUSTRIAN GRID
12/14
3 Milestone – Selection of pMHC-complex
Milestone Date Description
Focus 1a:
Selection of relevant pMHC-complexes
2005-03-31 pMHC complexes shall be identified which are of scientific/clinical relevance, for which 3D structural data and also TCR binding data exist or can be produced.
A MHC-antigen table was made up by U. Omasits. The table is based on the MPID (MHC-
Peptide Interaction Database 1.3), which lists all MHC-complex crystal-structures of the PDB
(The RCSB Protein Data Bank). These structural data are essential for any pMHC simulation.
All 90 entries were included in the table and additional data was taken from the PDB.
Furthermore binding-affinity data from the two versions of JenPep (Peptide Binding
Database; Version 1 and 2) were added to the table. This table should facilitate the further
selections of clinically relevant pMHC-complexes.
MPID – http://surya.bic.nus.edu.sg/mpid/index.html
PDB – http://www.rcsb.org/pdb/index.html
JenPep 1 – http://www.jenner.ac.uk/jenpep1/
JenPep 2 – http://www.jenner.ac.uk/jenpep2/
The human B*2705 MHC-class I allele (PDB-entry: 1HSA) was chosen for first simulations
because simulations with this pMHC-complex have already been performed.
AUSTRIAN G
RID
13/14
PDB C
ode M
HC type M
HC s
ourc
e M
HC a
llele
Rele
ase Y
ear
Peptide S
ourc
e
Peptide S
equence
H-b
onds JenPep (R) IC50 (nM
) JenPep t 1
/2 (m
in)
1HHJ
class I
Human
A*0201
1993
Synthetic
ILKEPVHGV
14
12 / 192 / 909
1AKJ
class I
Human
A*0201
1997
HIV-1 RT
ILKEPVHGV
13
12 / 192 / 909
1HHK
class I
Human
A*0201
1993
Synthetic
LLFGYPVYV
10
3.8 / 13
3000 / 6400
1AO7
class I
Human
A*0201
1997
HTLV-1 Tax
LLFGYPVYV
10
3.8 / 13
3000 / 6400
1BD2
class I
Human
A*0201
1998
HTLV-1 Tax
LLFGYPVYV
11
3.8 / 13
3000 / 6400
1B0G
class I
Human
A*0201
1998
Human-peptide P0149
ALWGFFPVL
12
1HHG
class I
Human
A*0201
1993
HIV-1 gp 120
TLTSCNTSV
12
1HHI
class I
Human
A*0201
1993
Synthetic
GILGFVFTL
9
12.4
1000 / 1200
1B0R
class I
Human
A*0201
1998
Influenza m
atrix
GILGFVFTL
7
12.4
1000 / 1200
2CLR
class I
Human
A*0201
1998
Synthetic
MLLSVPLLIG
10
1HHH
class I
Human
A*0201
1993
HBV nucleocapsid
FLPSDFFPSV
11
2.5 / 3.3 / 1.6 / 1.2 / 2.6 / 0.57
1TMC
class I
Human
A*6801
1995
Synthetic
EVAPPEYHRK
14
1AGB
class I
Human
B*0801
1997
HIV-1 gag
GGRKKYKL
15
1AGC
class I
Human
B*0801
1997
HIV-1 gag
GGKKKYQL
18
1AGD
class I
Human
B*0801
1997
HIV-1 gag
GGKKKYKL
16
1AGE
class I
Human
B*0801
1997
HIV-1 gag
GGRKKYKL
15
1AGF
class I
Human
B*0801
1997
HIV-1 gag
GGKKRYKL
14
1HSA
class I
Human
B*2705
1992
/N
ARAAAAAAA
14
1A1N
class I
Human
B*3501
1998
HIV-1 Nef
VPLRPMTY
11
1A9E
class I
Human
B*3501
1998
EBV-Ebna3c
LPPLDITPY
12
1A9B
class I
Human
B*3501
1998
EBNA-3C
LPPLDITPY
12
1A1M
class I
Human
B*5301
1998
HIV-2 gag
TPYDINQML
12
1VAC
class I
Murine
H2-Kb
1996
Ovalbumin
SIINFEKL
14
1.4
1VAD
class I
Murine
H2-Kb
1996
Yeast alpha glucosid
SRDHSRTPM
21
2VAA
class I
Murine
H2-Kb
1996
Vsv nucleoprotein
FAPGNYPAL
16
1CE6
class I
Murine
H2-Db
1999
SV nucleoprotein
FAPGNYPAL
15
1QLF
class I
Murine
H2-Db
1999
SV-nucleoprotein
FAPSNYPAL
13
1BII
class I
Murine
H2-Dd
1998
HIV-1 P18-100
RGPGRAFVTI
14
9.5
1LDP
class I
Murine
H2-Ld
1998
Natural peptide
APAAAAAAM
9
1I1Y
class I
Human
A*0201
2000
HIV-1RT
YLKEPVHGV
13
1I1F
class I
Human
A*0201
2000
HIV-RT
FLKEPVHGV
11
1DUZ
class I
Human
A*0201
2000
HTLV-1 Tax
LLFGYPVYV
11
3.8 / 13
3000 / 6400
An extract of the MHC-antigen table.
AUSTRIAN GRID
14/14
4 Milestone – Select graphics software
Milestone Date Description
Focus 1b:
Select graphics software
2005-05-01 Test molecular viewers and select appropriate products for a) high quality rendering and b) movie production. Establish data link to simulation results.
There are quite a few free molecular viewers available. VMD (Visual Molecular Dynamics)
will be used to visualize the Gromacs trajectories (the output file of a molecular dynamics
run) in high quality. RasMol will be used for lower quality preview of crystal structures or
stills from the molecular dynamics simulation.
VMD – http://www.ks.uiuc.edu/Research/vmd/
RasMol – http://bioinformatics.yale.edu/modeling/rasmol/rasmol.html
5 Milestone – Produce Graphics
Milestone Date Description
Focus 1b:
Produce Graphics 2005-07-01
Produce graphics of molecules during simulation (stills and movies) using appropriate molecular viewers.
U. Omasits produced a short movie-sequence of the preparation work to simulate a mutated
peptide using RasMol combined with own scripts:
The animation starts with the whole crystal structure and slowly changes to the ribbon
representation. Next, the lower part of the peptide (alpha-3-domain) is cleaved away in order
to save computational power. Finally, the antigen nonamer is outlined and a few amino-acids
are exchanged.
Links to external sources – like the more detailed report on “Installation and Test of
Molecular Dynamics Simulation Packages on SGI Altix and Hydra-Cluster at JKU Linz” by
DI Rene Kobler, the whole pMHC-table and the short animation-sequence – will follow soon!
Related Documents
