Main     Research     Members     Publications     Links     Positions     Contact


 

Research

 

The research interests of the Walker Molecular Dynamics lab are in the field of Molecular Biology and Computational Chemistry, in particular, in the development of advanced methodologies for the simulation of enzymes. Interests include a combination of both the development of the underlying algorithms, parameter sets and associated software as well as the application of these developments to study enzymes related to human diseases and efficiency in the production of bioethanol.

 

Acceleration of Molecular Dynamics Simulations using Graphics Processing Units


Modern GPUs offer a substantial amount of untapped processing capability for very little hardware or infrastructure cost.

In a joint collaboration with NVIDIA the Walker Molecular Dynamics lab are developing GPU accelerated versions of the major MD engines in the AMBER Molecular Dynamics Package. An initial version supporting implicit solvent Generalized Born calculations and explicit solvent Particle Mesh Ewald simulations in the NVE, NVT or NPT ensembles was released in April 2010 as part of the AMBER 11. Current NVIDIA cards offer speedups of between 20 and 200x over a single Intel Nehalem core. (AMBER on GPUs)
 

Work is continuing to further expand the set of features that are supported, to improve performance further and to provide support for acceleration across multiple GPUs within a single node or across multiple nodes. Using the latest generation of GPUs a single $2800 node is capable of running the JAC DHFR HMR NPT benchmark at over 300ns/day making this software the fastest molecular dynamics package on commodity hardware.

This work includes the conceptualization and development of the Life Sciences GPU Certified Computing progam. This program was developed initially in collaboration with Exxact as the AMBER Certified GPU Computing program. It has since been expanded to include other Life Sciences Software.

 
Custom Hardware and Software Solutions for Life Sciences and Machine Learning Applications

In close collaboration with industry leaders in accelerated computing including NVIDIA, Intel and Exxact Corp the Walker Lab has designed a number of optimized GPU and Xeon Phi computing solutions for scientific research in the Life Sciences and HPC "supercomputer in a desktop" systems for Deep Machine Learning and Convolution Neural Networks.

These custom designed systems are some of the most powerful desktop and server resources ever produced. For example the 4 GTX-Titan-X systems pack over 30 TFLoPs of computing power into a single desktop system. Custom designed motherboards support up to 20 GPUs in a single 4U node. That's 140 TFLoPs in a 4U node or 1.4 PFLoPs in a rack. Specific projects include:

i) AMBER certified GPU MD computing systems.

ii) Gromacs optimized GPU MD computing solutions.

iii) Exxact - Life sciences optimized GPU and Xeon Phi certified platforms.

iv) Turnkey NVIDIA Digits and deep learning neural network GPU accelerated development boxes. (NVIDIA Digits Dev Box)

v) Optimizing PCIe fabric design for optimum peer to peer communication in multi GPU machine learning and convolution neural networks.

vi) Relion preinstalled GPU accelerated Cryo-EM refinement solutions.

Research teams that utilize these Walker Lab and Exxact hardware and turnkey system designs include: University of California San Diego, Raytheon, Lawrence Livermore National Laboratory, Los Alamos National Laboratory, University of Georgia, John Hopkins University, Carnegie Mellon University, Revolution Medicine, University of Macau, James Madison University, Chapman University, University of Kentucky, Brooklyn College, Trinity University, Truman State University, Intel, University of Kansas, National Institutes of Health, St. Jude's Children's Research Hospital, Fujitsu, University of Houston, Memorial Sloan Kettering Research Center, Mayo Clinic, The College of New Jersey, Samsung Semiconductor, University of Akron, Talca University, Extreme Design Solutions Group, Lamar University, Harvard Medical School, University of Southern Denmark, UC Irvine, UC Riverside, College of Charleston, Hong Kong Baptista University, UT Brownsville, University of Southern California, EPFL, University of Geneva, Cedar Sinai Medical Center, University of Chicago, Virginia Tech, University of Denver, Green Energy Americas, Claratech, Rutgers University, Sanford Burnham Medical Research Institute, Sandia National Laboratory, Jordan University of Science and Technology, Novartis Institute for Biomedical Research, Pfizer, Dart Neuroscience, FDA, University of Tennessee...

Example from HPC Wire: Exxact GPU Cluster Accelerates Cancer Research at Notre Dame

For more info or to request a quote for your own customized GPU computing solution please contact ross at rosswalker.co.uk.

 
Development of Advanced QM/MM Techniques for Enzyme Simulation

One of the primary limitations of classical molecular dynamics simulations is the inability to break covalent bonds during a simulation. This prohibits the direct observation of reaction events during a simulation. In addition classical MD simulations require parameters for all bonded and non-bonded interactions within the system. Such parameterization is typically done for all amino and nucleic acids but a significant number of chemically interesting proteins employ co-enzymes and catalytic metal centers which are typically not covered by traditional protein force fields.
 

We are developing advanced, hybrid quantum mechanical/molecular mechanical (QM/MM) techniques that offer a possible solution for these limitations. Research in the lab currently focuses on:

1) Improvements in the accuracy of semi-empirical Hamiltonians.

2) Links between AMBER and ADF to support DFT based QM/MM simulations.

3) Performance and parallel scaling improvements to QM/MM MD simulations.

4) Development of techniques for allowing variable QM regions during MD simulations.

5) Standalone libraries for high performance semi empirical QM calculations.

 

Development of Advanced Biofuels from Cellulosic Biomass


One of the biggest challenges that mankind will have to face over the next 20 years is the transition from energy sources based on fossil fuels to efficient carbon neutral renewable energy technologies. There is no simple solution to this problem and instead a plethora of methods will need to be developed and employed. One promising technology is the conversion of cellulosic plant matter into bioethanol that can subsequently be used as a replacement for gasoline.
 


Image courtesy of the National Renewable Energy Laboratory

In collaboration with the National Renewable Energy Laboratory, Cornell University and the University of Michigan and funded by the Department of Energy Scientific Discovery through Advanced Computing (SciDAC) program we are developing advanced techniques for studying the enzymatic degradation of cellulose as part of research to improve the efficiency of bioethanol production.
 

A Comprehensive Phospholipid Membrane Force Field


Phospholipids serve a major function in the cells of all organisms forming the cell membranes of all living things. Increasingly more and more is being learnt about the structure and function of membrane bound proteins and yet there exists only minimal classical force field support for phospholipid membranes.

In collaboration with Prof. Ian Gould at Imperial College London and Prof Knut Teigen at the University of Bergen, Norway, we are developing the next generation of phospholipid force fields for MD simulations. The are being designed in such a way that any combination of phospholipid membrane can be built and simulated. The latest version, Lipid 14 was released in April 2014 and is the first modular all-atom lipid force field to support unbiased NPT simulation. It thus does not require artificial restraints such as constant surface tension. Additionally this force field is the first all atom lipid force field that demonstrates self assembly of lipid bilayers from random mixtures of phospholipids in water.


Image of the enzyme Rhodopsin embedded in a membrane bilayer.

 

Predicting Enzyme Activation Pathways


Traditionally drug discovery has focused on finding ligands that bind to enzyme active sites and inhibit their reactivity. However, there exists plethora of alternative approaches that seek to inhibit enzyme activation.

One such example is the Adenovirus protease enzyme which is common to a number of viruses including those responsible for avian and swine flu. As part of the final stage of virus replication the virus must remove the chemical scaffolding that was constructed as part of the replication process in order to break out of the host cell and infect new cells. This is achieved by the adenovirus protease enzyme which is initially synthesized by the virus in an inactive form and must then undergo an elaborate activation process. Hence the development of drugs to inhibit this pathway is a major research aim.

We are developing both the methods, such as the Nudged Elastic Band (NEB) approach, and applying these, in collaboration with Brookhaven National Laboratory, to determine the activation pathways and possible inhibition sites for the Adenovirus Protease. This work is uncovering intermediate states along the adenovirus activation pathway that could ultimately lead to the development of an entirely new class of antiviral drugs.


 

 


the Walker Molecular Dynamics lab