(Updated on 2013/3/4)
ID:M-1
Principal developer
Akinori KIDERA, Team Leader, Molecular Scale R&D Team, Computational Science Research Program (CSRP), RIKEN
General description
Platypus-MM/CG targets biomolecules, such as proteins, DNAs, and drugs. It aims to simulate the interactions between biomolecules and the involved 3D conformational changes, whose time scale is too long (msec to sec) to be solved with the conventional MD methods.
Computational model
Multi-copy, multi-scale coupled molecular dynamics (MD)
Computational method
Replica-exchange method, Multi-scale essential sampling method (MSES), On-the-fly string method, and Particle filters
Parallelization
MPI among copies, and MPI-thread hybrid parallelization for the divided loops
Required language and library
C++, MPI, OpenMP, LAPACK, FFTW, and NetCDF
Status of code for public release
Source code is available through ISLIM download site.
Maximum computing size in present experience
- 3D conformational changes and pathway searches for adenylate kinase
- 16-million-atom system (60,000-atom system x 256 copies)
- 8,192-core parallel computing with RIKEN RICC system
- Required memory/disk storage: 52 GB/10 TB
Expected computing size in K computer
- Multi-drug efflux transporter AcrB
- 1-billion-atom system (500,000-atom system x 256 copies)
- Parallel computing with 65,536 cores (256 cores/copy)
- Required memory/disk storage: 3.3 TB/1 PB.
Figure 1. The concept of on-the-fly string method (upper), and the application results for a protein (lower).
What does the code enable?
- Simulation of processes for biologically active agents like drug to dock biomacromolecules like protein
- Prediction of 3D conformational changes in docking for drugs and in coupling of proteins
- Clues leading to discovery of bioactive materials or function modified proteins based on above simulations.
ID:M-2
Principal developer
Yuji SUGITA, Associate Chief Scientist, RIKEN Advanced Science Institute
General description
The code computes multidimensional replica exchange MD to efficiently predict protein conformations with surrounding environment, conformation change, free energy values. The code uses MARBLE, which will run on the K computer, for MD calculation. Other MD codes can be applicable to Platypus-REIN
Computational model
Classical particles (MD)
Computational method
Replica exchange method
Parallelization
Parallel execution of multireplica
Required language and library
Fortran90, C, MPI
Status of code for public release
Source code is available through ISLIM download site.
Maximum computing size in present experience
(AAQAA)3 protein (10-thousand-atom system) run on RICC's 8,192 cores
Expected computing size in K computer
- Protein's conformation change and conformation prediction.
- Binding free energy profile between proteins
- Analysis of 1-million-atom system with 640 thousands of cores.
Figure 1. Analysis of a protein by the replica exchange method.
What does the code enable?
- Predicting the conformation or conformation change for membrane proteins and unprecedented largest proteins
- Deeper understanding about the mechanism for forming protein complexity in the atomic scale resolution
- As an application, the deep understanding about protein-disease relationship can help medicine and drug discovery in the future.
ID:M-3
Principal developer
Mitsunori IKEGUCHI, Associate Professor, Supramolecular Biology, Yokohama City University
General description
The code performs all-atom biomolecular simulations for protein, DNA, and membranes.
Computational model
Classical all-atom molecular dynamics simulations
Computational method
- Particle Mesh Ewald method
- Symplectic numerical integration method
Parallelization
Domain decomposition. Thread parallelization
Required language and library
C, MPI, OpenMP
Status of code for public release
Source code is available through ISLIM download site.
Maximum computing size in present experience
- A 100-thousand to 1-million-atomic system
- Cray XE6's 3,000 cores
- Required memory/disk storage: 4 GB per node/1 TB
Expected computing size in K computer
- Multidrug efflux transporter (about 500 thousand atoms)
- Iterations: >109
- Required memory/disk storage: 16 GB per node/10 TB.
Figure 1. Multidrug efflux transporter AcrB
What does the code enable?
- Simulations for large biomolecules with explicit membrane and solvent molecules
- By using simulations of the multidrug efflux transporter that causes drug resistance, the molecular mechanism underlying multidrug efflux will be elucidated.
ID:M-4
Principal developer
Shoji TAKADA, Associate Professor, Kyoto University
General description
Using the coarse-grained molecular model, the code does a long simulation of large biomolecules
Computational model
The coarse-grained classical MD model
Computational method
Time integration of the Langevin equation (differential equation)
Parallelization
Neighbor list approach, replica exchange
Required language and library
Fortran90, MPI, OpenMP
Status of code for public release
Source code is available through ISLIM download site.
Maximum computing size in present experience
- A several-millisecond simulation for the 10 thousands of residues of the protein
- Use 8,192 cores of the x86 cluster
- Required memory/disk storage: 2 GB per core/1 TB
Expected computing size in K computer
- A simulation for the process of milliseconds to seconds of the 100 thousands of particles (1 million atoms)
- Number of iterations: >1010
- Required memory/disk storage: 2 GB per core/1 PB.
Figure 1. A transporter (left) and a DNA-histone complex (right).
What does the code enable?
- The MD simulation for the dynamic behaviors of the proteins and DNAs for which structural data are available via the X-ray diffraction or NMR. In particular, the dynamic behaviors of protein complex, molecular motors, and transporters that exhibit large conformation changes can be simulated
- The typical examples are the dynamic behaviors for the kinesine motor, the multidrug efflux transporter, and the DNA-histone complex.
ID:M-5
Principal developer
Fumitoshi SATO, Professor, the University of Tokyo
General description
- The code solves the all-electron canonical wave functions of a protein, which requires complicated, large scale, and highly accurate computation, based on the density-functional approach.
- The code can do the first principle MD computation including excitation states.
Computational model
The molecular orbital (MO) method
Computational method
The Roothaan-Hall minimization (diagonalization) in the canonical MO basis by the direct method
Parallelization
RT algorithm + shell-type-classification equal-partitioning method
Required language and library
C++, MPI, OpenMP, ScaLAPACK
Status of code for public release
Maximum computing size in present experience
- The world-largest 306 residues 27,000 orbitals ground-state computation (Fig. 1)
- With Altix3700(64 CPU, 300 GFlops)
- Required memory/disk storage: 256 GB/1 TB
Expected computing size in K computer
- 20,000 atoms calculation in the ground states, which covers the most proteins
- Dynamic analysis
- Excitation states
- Required memory/disk storage: 2 PB/10 PB (Excitation state).
Figure 1. The differece electrostatic potential in the insulin hexamer between the ProteinDF and a classical computation.
What does the code enable?
- The code contributes to not only the highly reliable basic research in the drug discovery but also more highly qualified and efficient drug development as well as the creation of next-generation R&D models in the pharmaceutical.
- It can bring the Japan's superiority in patents over foreign countries.
- It can apply to catalysts, molecular elements, and environmental materials.
ID:M-6
Principal developer
Shigehiko HAYASHI, Associate Professor, Kyoto University
General description
The code determines the free-energy optimal conformation in the reactive subtrate molecule in the biomolecules, and analyzes its reactivity by the hybrid approach with the quantum chemistry method (QM) and the molecular field method (MM).
Computational model
The molecular orbital method, the molecular force field method
Computational method
Reweighting method (Free energy), Ewald method (Coulomb)
Parallelization
Conformation sample decomposition for MM
Required language and library
FORTRAN77, the socks library in GAMESS
Status of code for public release
To be released.
Maximum computing size in present experience
- The free-energy optimal conformation for polysaccharide (dimer portion) of the a-amylase kinase
- A 70-thousands-of-atoms system, QM atoms: 69, base functions: 650, MM conformation samples: 140,000
- Required memory/disk storage: 256GB/50GB (128 cores)
Expected computing size in K computer
- Multidrug efflux transporters
- About 500 thousands of atoms, 1-million MM-conformation samples
- Required memory/disk storage: 4 TB/ 4 TB.
Figure 1. The transition state in the hydrolysys reaction of a polysaccharide subtrate in the a-amylase kinase.
What does the code enable?
- The reactivity analysis for a subtrate in the kinase reactions
- Understanding of the molecular function driven by the proton concentration gradient, which is the most important function for the biological energy conversion.
ID:M-7
Principal developer
Nakamura HARUKI, Professor, Institute for Protein Research, Osaka University
General description
The first principle calculation for all-electrons of the molecule such as protein by the molecular orbital (MO) method and DFT theory
Computational model
MO, DFT theory, Hybrid DFT with Hartree-Fock
Computational method
Quadrature of the multicentered Euler-Maclaurin-Lebedev, eigenvalue problem, and so on
Parallelization
MPI(to be extended to the hybrid parallelization)
Required language and library
Fortran77, Fortran90, MPI, HDF, ScaLAPACK, LAPACK, etc
Status of code for public release
Source code is available through ISLIM download site.
Maximum computing size in present experience
- Pigments like bacterio-chlorophyll
- Several-hundred to one-thousand-electrons system with 8,192 cores
- Required memory/disk storage: 1.2GB per core/100 MB
Expected computing size in K computer
- The electron structure at the ground state and the excitation states for a several-hundred to several-thousands-of-electrons system with 640 thousands of cores
- Required memory/disk storage:1.6GB per core/several GB.
Figure 1. The homo orbital for the pigment (bacterio-chlorophyll dimer) at the photosynthetic reaction center derived from Rhodobacter sphaeroides.
What does the code enable?
- Understanding of the electron structure at the ground and excitation states for a protein system that contains transition metal
- Understanding of the reaction mechanism by the electron structure analysis
- The parameter calculations, such as the electromagnetic interaction, given a probe like the Electron Paramagnetic Resonance (EPR), for unknown protein structure.
ID:M-8
Principal developer
Nakamura HARUKI, Professor, Institute for Protein Research, Osaka University
General description
The code gives the precise analysis for a biological polymer such as a kinase reaction mechanism by introducing the electron state effect to the molecular mechanics (MM) approach
Computational model
Molecular orbital method, molecular dynamics (MD) method
Computational method
The direct solving method for dense matrix diagonalization, the iterative solving method for sparse-matrix simultaneous equations, an improved Wolf method, etc
Parallelization
Domain decomposition (MO), atom dividing (MD)
Required language and library
FORTRAN77, Fortran90, C, MPI, OpenMP, ScaLAPACK, LAPACK
Status of code for public release
To be released.
Maximum computing size in present experience
- The isomerization's thermal mechanical process for proline and N-metylacetamide (quantum system) in the water (classical treatment for H2O)
- Twenty thousands of atoms by a 256-core x86 cluster
- Required memory/disk storage: 300MB per core/3GB
Expected computing size in K computer
- The reactive free-energy map for a several-hundred to several-thousands-of-electrons system surrounded by several hundred thousands of the classical atoms using 640 thousands of cores
- The dynamics analysis for both the ground and excitation states
- Required memory/disk storage: 1.6GB per core/several PB.
Figure 1. Coupled quantum-classical calculation for the reaction coordinate dependency of pyramidality.
What does the code enable?
- Useful knowledge for the kinase reaction process being important for life and the drug development controlling such reactions.