All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
Molecular Simulation to build models for enzyme induced fit
1. 분자 설계 연구실
홈페이지 http://www.bmdrc.org/
전화 02-393-9550, 이메일 mskim@bmdrc.org
연세대학교
세미나3 발표
- Molecular Simulation to build models
for enzyme induced fit
2015년 5월 22일(금)
조교 : 김 민 성 (통 합)
지도 교수 : 노 경 태 교 수
2. Lab : 공학관B408
Phone : 02-2123-7739
Mail : ktno@yonsei.ac.kr
Homepage : http://www.csblab.or.kr
Computational Systems Biology Lab.
Professor : Kyoung Tai No
Computational Chemistry Cheminformatics
Solvation free energy, charge
model, and forcefields; gives
concrete understanding and
analysis tool for further
developments.
Statistical analysis of
multivariate chemical
feature space via machine
learning techniques.
Spectral similarity
Structural similarity
Activity similarity
Natural Product Profiling&Networking
Profiling natural product/metabolite in high-throughput manner,
revealing its biological activity.
Commercial
available screening
Database
(12 Millions)
PPI Screening
Library
Development of
PPI focused screening library
(0.2 Millions)
Target-focused Library
Design
Pharmacophore Based Screening
Structure-based
Pharmacophore screen
Screen of protein
Interaction surface
Ligand-based
pharmacophore screen
Virtual screening
Virtual hits
Predicted binding mode
•ASN159 Hot Spot region
•GLU196 Hot Spot region
ASN159
GLN160
LYS154
GLU196
“Hotspot” binding region:
•Define binding site
•Hydrogen bond region
Biding Site Prediction
Flora Genesis System In silico Drug Design
RESEARCH INTEREST
3.
4.
5. What is molecular Dynamics
• A computational microscope
• An experiment on a computer
• A simulation of the classical mechanics of atoms
6. GOOD Energy Calculation Function, Force Fields, for DGsystem
GOOD Simulation Method for DGsystem
t1 t2 t3 t4 t5 t6 t7 t8 tn tn+1 tn+2 tn+3
S1 S2 S3 S4 S5 S6 S7 S8 Sn Sn+1 Sn+2 Sn+3
G1 G2 G3 G4 G5 G6 G7 G8 Gn Gn+1 Gn+2 Gn+3
Energy/Mechanics
Based Design
Time
Structure
Free Energy
7. Systems in a Life System
Atom
10-12 m
Protein
10-9 m
Cell
10-6 m
Tissue
10-3 m
Organ
100 m
Organ System
& Organism
Physiology
Gene
Networks
Pathway
Models
Stochastic
Models
Differential
Equation
Continuum Model
Partial Diff. Eqn
Systems
Model
10-6s
Molecular Events
ion channel gating
10-3s
Diffusion
Cell signaling
100s
Mobility
103s
Mitosis
104s
Protein
Turnover
109s
Human
Lifetime
Spatial and temporal levels encompassed by biological systems
10. Speed Isn’t Everything
• How accurate are molecular mechanics force fields?
- Clearly good enough for some biologically and
pharmacologically important applications
• Where are the weak points?
- Polarizability? Hydrogen bonds? Combinning rules?
• Can we Improve the accuracy of today’s force fields?
- At what cost in execution speed?
• Even negative results could provide biologically and
pharmacologically relevant insights
11. Research Interest Areas
Force Fields
SBFF
CHARM
AMMBER
MMFF
Simulator
Lammps
Gaussian
Schrödinger
Application
PPI
Molecular Modeling
Eco Engineering
Appropriate Tech
Consilience Approach
14. 2
0 )( ddkEstretch
)( 0
1 dd
estretch eDE
2
0 )( kEbend
)cos(1 S
S
S
kEtorsion
Intra Molecular Motions and their PEF
15. Classification of Force Fields
Classical FFs
AMBER, CHARMm, CVFF, ECEPP/2, Homans’ FF, Pullman (DNA)
Second-generation FFs
CFF91, PCFF, CFF95, MMFF93, MM
Water potential models (Flexible or non-flexible, inclusion of ploarization or not
ST2, TIP3, TIP4, SPC, CVFF, OPLS
COSMO, FDM, BEM, SMx, SFED
Broadly applicable FFs
UFF, Dreiding FF, ESFF (Extensible systematic FF), ……..
Special-purpose FFs
Glass, Zeolite, sorption, Morphology, ……..
16. For Small Organic molecules
MM2
- For structure determination of small organic molecules
- Developed by Allinger at U. Georgia
- FF parameters are determined with spectroscopic data
MM3
- Accurate vibrational frequency than MM2
MMFF93: Merck Molecular FF
- Using QM calculation as constraints for FF parameters fitting
Tripos force field
- For small organic molecules
Classification of FFs; Small Organic
Molecules
17. CFF : Consistant FF (CFF91, PCFF, CFF95)
- Contains both Anharmonic term and cross terms
CFF91- Hydrocarbons, Proteins, protein-ligand
PCFF- Polymer, organic materials
CFF95-Biomolecules, organic polymers
- for small organics and liquid and solid simulations
Shortcomings of above force fields
- inadequate for inter molecular interaction
- does not include electrostatic interaction
- van der Waals radii are too small
Classification of FFs; Small Organic
Molecules
18. ECEPP, ECEPP/2 SBFF (Self Balance Force Field)
Protein structure, in torsional space (no stretching & bending)
Harold Scheraga at Cornell U, Kyoung Tai No at Yonsei U
AMBER (Assisted Model Building with Energy Refinement)
Protein / Nucleic Acids, Peter Kollman at UCSF
CHARMM (Chemistry at HARvard using Molecular Mechanics)
Mainly for Protein, Martin Karplus at Havard
GROMOS (GROenigen MOlecular Simulation)
van Gunsteren and Berendsen at ETH Zurich.
CVFF(Consistent-valence force field )
Dauber-Osguthorpe, out-of-plane energy calculation included
For amino acids, water, and a variety of other functional groups
Classification of FFs; Biomolecules
19. Dreiding force field,
1st and 2nd period elements
Goddard at Caltec / Mayo at Biodesign / Olafson
UFF (Universal Force Field)
Include most of elements in Periodic table
Rappe at Colorado State U. / Casewit at Calleo Scientific /
Skiff at Shell Research
Classification of FFs; Broadly Used
20. Self-Balanced Force Field (SBFF)
• 1) accurate intra- and
inter-molecular Potential
Energy Function (PEF),
and
• 2) good simulation
algorithm that describes
nature of the molecular
worlds.
23. 23
Lammps code is Object oriented which is very similar to JAVA
Lammps has a huge diversity of force-fields you can use,
and also you can define new force-fields.
Which makes it seems good candidate for BMDRC
Object Oriented Lammps
26. 26
This example script shows how to run an atomistic simulation of uniaxial
tensile loading of an aluminum single crystal oriented in the <100>
direction.
Practice 4 : Aluminum Uniaxial Tension
Data retrieval was denied
due to Dr. Raju’s
Calculation
33. 33
Result - Peptide solvation
72.105
74.11
1 THREAD 4 THREAD
timesteps/s
Comparison of serial & parallel calc
Loop time of 346.715 on 1 procs
for 25000 steps with 2004 atoms
99.0% CPU use with 1 MPI tasks x
1 OpenMP threads
Performance: 0.019 ns/day
1284.129 hours/ns 72.105
timesteps/s
Loop time of 337.334 on 4 procs
for 25000 steps with 2004 atoms
99.2% CPU use with 1 MPI tasks x
4 OpenMP threads
Performance: 0.019 ns/day
1249.386 hours/ns 74.110
timesteps/s
34. 34
High Performance Computing(HPC) Cloud Platform
• Key Cloud Properties
• Cloud HPC: Good & Evil
• Success Stories
• Features & Opportunities
39. 39
High Performance Computing(HPC) Cloud Platform
• Key Cloud Properties
• Cloud HPC: Good & Evil
• Success Stories
• Features & Opportunitie
40. 40
High Performance Computing(HPC) Cloud Platform
• 1. Works even with limited budget.
• 2. Perfect for infrequent “monster” jobs.
• 3. Helps to reduce Time to Market.
• 4. Enables disaster-resiliency.
• 5. Reduces IT complexity.
The Bright Side
• 1. Performance and latency issues.
• 2. Data volume issues.
• 3. Vendor-related issues.
• 4. Security concerns.
• 5. Cost-effectiveness concerns.
The Dark Side
42. 42
High Performance Computing(HPC) Cloud Platform
We probably don’t want to use the Cloud
if we have this*:
* however, the costs might not be our primary concern
43. 43
High Performance Computing(HPC) Cloud Platform
• Key Cloud Properties
• Cloud HPC: Good & Evil
• Success Stories
• Features & Opportunitie
44. 44
High Performance Computing(HPC) Cloud Platform
• Batch Processing: New York Times and MapReduce
- 4 TB raw images, 11M PDFs, 100 Hadoop workers = $240
• Data Processing: Morgridge Institute for Research, gene indexing
- 1M core-hours, high-memory EC2 Spot instances: < $20K paid
• Simulations and Analysis: Schrödinger (drug research)
- 50K cores, 21M chemical compounds: < $5K paid
- (Amazon infrastructure value estimated at $20~40M)
HPC Cloud Case Studies
45. 45
• Simulations and Analysis: Schrödinger (drug research)
High Performance Computing(HPC) Cloud Platform
48. 48
Novartis Uses AWS to Conduct 39 Years of Computational Chemistry In 9 Hours
High Performance Computing(HPC) Cloud Platform
http://youtu.be/oa-M9GcaDN0
49. Molecular modeling is the study of the geometry and properties of
molecules by computer-aided techniques.
Molecular modeling is a growing area in science & technology to
explain the phenomena at molecular level.
visualize molecules
study the structure of molecules
study the properties of a molecule
compare the structure and properties of molecules
study interactions between molecules
study reaction mechanisms
predict the structure of molecules
predict the properties of molecules
predict reaction mechanisms
Molecular Modeling (Design)
50. Computer Aided Drug Design
Receptor Structure
Unkonwn
Receptor Structure
Konwn
Ligand
Structure
Unknown
Combinatorial
Chemistry
3D Structure
Generation
De novo Design
Receptor Based 3D
Searching
Ligand
Structure
Known
Pharmacophore
Define
QSAR
Structure Based
Optimization
Affinity Calculation
51. Computer Added Molecular Design
Energy – Mechanics Based Design
Research
Subject
Mechanics
Based Model
Simulation
X, T, t, E
Expiments
Analysis
Prediction
QM, E-FF, TD / MM, MC, MD
Knowledge Based Design
DATA Analysis RULEs Prediction
QSAR
13.