Acellera's Blog

Adaptive Sampling for MD Simulations Talk at GTCBio

by David Soriano

A few of weeks ago we were invited to present at GTCBio Drug Design and Medicinal Chemistry 2014 Conference. The conference was very interesting and very well organized, and we will make sure to keep it our agenda for future years. The meeting also happened to be in Berlin, which instantly became one of my favorite cities. Our talk was titled “Machine Learning in FBLD” and discussed our new adaptive sampling method for probing protein ligand interactions using atomistic molecular dynamics simulations that was developed in collaboration with the De Fabritiis lab. So how did we get there?

High-throughput molecular dynamics can be used for profiling protein-ligand interactions

In 2011 we reported our seminal work towards a practical approach for profiling protein ligand interactions in silico with all atom resolution. In this proof-of-principle study we showed that with enough sampling we could use atomistic high-throughput molecular dynamics simulations to reconstruct the binding of benzamidine to trypsin, and not only obtain accurate energies, kinetics, poses but also resolve a ligand binding pathway. We were also able to predict several metastable states later observed experimentally.

Application of molecular dynamics simulations in fragment based lead discovery (FBLD)

We next focused on expanding the library size, and on adapting the methodology to fragment based ligand design. Benzamidine is a small molecule of micromolar affinity for trypsin, and therefore extending this method to FBLD was a natural progression. For the fragment study we used a 2003 STD NMR study that targeted Factor Xa, a target related to trypsin, and a set of forty fragments. Once more our MD experiments were able to recapitulate the binding of the library to the target, and also give accurate representations of binding energies, kinetics, poses and binding pathways. Importantly, we were also able to rank the compounds in order of increasing binding affinity accurately. As exciting as the results were, we needed to collect 2milliseconds of aggregate biological timescale trajectory data, a challenge we could only meet because of our access to GPUGrid through our collaboration with the De Fabritiis group.

Development of an adaptive sampling protocol for MD

One limitation of our HTMD based in-silico binding method at that time involved the amount of sampling needed for system convergence, which was in the order of 50×10^-6 s per compound for Factor Xa. Clearly, in order to make this high-throughput molecular dynamics methodology a practical complement to current experimental FBLD screens, we needed to improve the efficiency of our simulation methods. To this end, we developed a fully automated adaptive sampling method that can deliver system convergence about an order of magnitude faster than the brute-force HTMD approach.

As opposed to HTMD, where we run hundreds of simulations at once, our adaptive sampling protocol begins with only ten trajectories each starting from random initial states and each a few tens of nanoseconds long. We then analyze these trajectories using Markov State modelling, and use residence time information obtained from clustering to select the starting conformations of future runs. This run-analyze-respawn cycle is defined as an epoch, and each adaptive experiment is composed of ten epochs. Note that all of this is automated and that we only need to setup the initial trajectories before working up the results after the last epoch. For each protein-ligand system we run ten replicate adaptive experiments.

For our proof-of-concept we tested this adaptive sampling method on trypsin-benzamidine, a system for we which we have a very large amount of data and that over the years has become our benchmark. As opposed to regular HTMD where we needed to collect about 50×10^-6 s of data for convergence, only 5×10^-6 s were needed in our adaptive experiment. In each experiment, learning was evident after the 3rd epoch, and the energy and pose matched those of our control HTMD experiment after the 5th epoch, in about 80% of the experiments.

What next?

We are very excited about these results as in principle — and assuming there were no other bottlenecks — we should be able to screen 400 compounds in the same amount of time previously spent with 40. This would make this technique fast enough for medicinal chemists to start looking at it as a practical and reliable complement to current FBLD screens. We hope to be there soon but we still need to test the universality of the method thoroughly as well as work out other kinks such as the issue of ligand parametrization.

gianniAdaptive Sampling for MD Simulations Talk at GTCBio
Read more

Atomistic MD Simulations and Experiment Provide Basis for Protein Motility in Prestin

Post by Mattia Sturlese, PhD.

Recently, we derived the structure of the prestin transmemebrane domain by combining bioinformatics, homology modeling, and MD simulations with extensive experimental work. Prestin (SLC26A5), a member of the SLC26/SulP anion transporter family localized in the outer plasma membrane of the outer hair cells (OHCs), is a motor protein essential for auditory processing. Motor proteins are a fascinating class of proteins able to transform molecular events into movements.

Prestin, a highly responsive piezoelectric transducer of unclear mechanism of action

Classical molecular motors are driven by ATP hydrolysis while prestin acts through a chlorine dependent intrinsic voltage sensor (IVS). In prestin, intracellular binding of a chloride ion induces a very large structural rearrangement – determined by the membrane potential – that endows this motor protein with a motility-related charge movement or nonlinear capacitance (NLC). One of the most attractive peculiarities of this mechanism relies on time scale in which this molecular event occurs, as it is remarkably faster than that of other cellular motor proteins. In a nutshell, prestin is a piezoelectric transducer able to convert membrane potential into macromolecular movement extremely efficiently. In fact, the piezoelectric coefficient in OHCs (20 fC/nN) is four orders of magnitude greater than found in man-made crystals used in electronic devices or manipulators. Despite its established physiological role as a key element in the high acuity of mammal hearing through cochlear amplification, it is essentially unknown how prestin can generate mechanical force. To date there are no known structures for the SLC26/SulP superfamily of anion transporters.

Computational modeling towards a predicted 3D prestin structure

Sequence-based bioinformatic analyses indicated the NCS2 and SulP transported families as possible candidates for homology modeling. Using hidden Markov models (HMMs) and their profile analysis we identified the uracil transporter UraA (PDB: 3QE7) as the best and most meaningful template for homology modeling. In order to validate the biophysics of this model we collected 150ns of atomistic molecular dynamics simulations using ACEMD. The MD simulation data showed that although there were defined local differences between the two models the two conserved the global fold and topology of UraA at equilibrium, thus corroborating the validity of our model. Thus, the prestin transmembrane domain appeared to be composed of 14 transmembrane segments organized into a 7+7 inverted repeat fold. Appropriate 3D structures to be used in further analysis were obtained by cluster selection.

Experimental validation of the model and understanding of the molecular basis for prestin’s activity

We continued validating the topology of the homology model using a substituted Cys accessibility method (SCAM).The SCAM results showed a notable agreement with water-protein contacts as derived from MD simulations. Importantly, the pattern of accessibility is in full agreement with the predicted topology (Fig 1), corroborating the model and UraA as an appropriate template for modelling SLC26. Solute accessible pathways were identified by analyzing the internal solvated areas and calculating the residence probability of explicit water molecules during the last 50 ns of the MD simulation. The central cavity is readily accessible from the intracellular space but occluded from the extracellular space by a constriction that precludes permeation from the exterior to the central cavity. A SCAM scan along the entire TM10 provided experimental confirmation of the model results; amino-acid positions predicted to contribute to the central cavity and its intracellular entrance were accessible to intracellular, but not extracellular MTS reagents. The accuracy of the model and the information from the MD simulations has also allowed identification of a candidate central binding site validated experimentally though mutagenesis.

Mattia Sturlese

Reference and further reading:

*Gorbunov, D., *Sturlese, M., Nies, F., Kluge, M., Bellanda, M., Battistutta, R., and ^Oliver, D. (2014). Molecular architecture and the structural basis for anion interaction in prestin and SLC26 transporters. Nat Commun 5.

Dallos, P., and Fakler, B. (2002). Prestin, a new type of motor protein. Nat Rev Mol Cell Biol 3, 104–111.

Oliver, D., He, D.Z.Z., Klöcker, N., Ludwig, J., Schulte, U., Waldegger, S., Ruppersberg, J.P., Dallos, P., and Fakler, B. (2001). Intracellular Anions as the Voltage Sensor of Prestin, the Outer Hair Cell Motor Protein. Science 292, 2340 –2343.

Zheng, J., Shen, W., He, D.Z.Z., Long, K.B., Madison, L.D., and Dallos, P. (2000). Prestin is the motor protein of cochlear outer hair cells. Nature 405, 149–155.

gianniAtomistic MD Simulations and Experiment Provide Basis for Protein Motility in Prestin
Read more

Supervised Molecular Dynamics (SuMD) as a helpful tool to depict GPCR-ligand recognition pathway in a nanosecond time scale

by Davide Sabbadin, PhD.

Brief abstract: Supervised Molecular Dynamics (SuMD) is a computational technique for investigating unbiased receptor-ligand recognition pathways, in the nanosecond timescale, by combining molecular dynamics simulations and a tabu-like algorithm.

G protein-coupled receptors (GPCRs) are transmembrane proteins that play a crucial role in linking various extracellular inputs with diverse cellular responses. It has been estimated that GPCRs constitute the target of about half of all drugs in clinical use today. Thus, from structural and pharmacological perspectives, represent an ideal target to design molecules with potential therapeutic effect (1). Understanding the GPCR-ligand recognition process is crucial for the development of drug candidates with favorable pharmacodynamic profiles.

Recent technological innovation has allowed Molecular Dynamics (MD) simulations to reach timescales comparable with those on which most biomolecular events of interest take place, thus closing the gap between theoretical models and experiments (2,3). Although now possible, it remains expensive to use unbiased Molecular Dynamics to quantitatively characterise ligand-protein binding.

Supervised Molecular Dynamics (SuMD) addresses this shortcoming, enabling the complete ligand-protein recognition process to be elucidated, using unbiased all-atom Molecular Dynamics, in a reduced timescale. The general method is not sensitive to the ligand’s starting position, chemical structure and receptor binding affinity (4).

As shown in the figure below, SuMD uses a standard MD simulation in which the ligand-receptor docking pathway is supervised by a tabu-like algorithm. During the production of the MD trajectory the distances between the center of masses of the ligand atoms and the residues composing the orthosteric binding site of the GPCR (dcmL-R) are monitored over a fixed time window (Δtck, e.g. 200 ps). An arbitrary number of distance points (n: a, b, c, d, e) per each checkpoint trajectory are collected and a linear function f(x)=m·x is fitted on the distance points at the end of the checkpoint time.

The tabu-like algorithm is applied to increase the probability of ligand-receptor binding events occuring without introducing bias to the MD simulation. More precisely, if m<0, the ligand-receptor distance is likely to be shortened over the checkpoint time and so the MD simulation is restarted from the last produced set of coordinates. Otherwise, the simulation is restored from the original set of coordinates and random velocities drawn from the Boltzman distribution are reassigned. This has the effect of causing the restarted simulation to explore a different region of configuration space, while maintaining sampling from the correct NVT ensemble. The tabu-like supervision algorithm is repeatedly applied until ligand-receptor distance (dcmL-R) is less than 5 Å.

Supervised Molecular Dynamics (SuMD) can be fully integrated into
ACEMD and enable investigating the complete binding process, in a nanosecond time scale, and to reproduce with high accuracy the crystallographic pose of protein-ligand complexes. Moreover, using SuMD simulations, it is possible to easily determine and characterize all possible metastable sites on the binding pathway to the orthosteric pose.

For more information visit:
Molecular Modeling Section from the Department of Pharmaceutical and Pharmacological Sciences, University of Padova

Figure: High affinity antagonist ZM241385 (yellow spheres) recognition pathway, using Supervised Molecular Dynamics (SuMD), to the membrane embedded human A2A Adenosine Receptor. During classic Molecular Dynamics (MD) simulations, the receptor-ligand distance vector (dcmL-R) is monitored and supervised in real time by a tabu-like algorithm. This computational method allows investigating the multifaceted unbiased ligand binding process in a reduced time scale of orders of magnitude, while taking advantages of the all-atom MD simulations accuracy of a GPCR-ligand complex embedded into explicit lipid-water environment.

Video: ZM241385-human A2A Adenosine Receptor recognition mechanism using Supervised Molecular Dynamics (SuMD). Simulation time in nanoseconds is reported. Van der Waals spheres represent ZM241385 atoms. Ligand nitrogen atoms are depicted in blue, oxygen atoms in red while carbon atoms are colored in white. Receptor ribbon representation is viewed from the membrane side facing transmembrane domain 6 (TM6) and transmembrane domain 7 (TM7). Hydrogen atoms are not displayed. Ligand binding sites are highlighted by arrows.

References:
(1) Jacobson, K. A.; Costanzi, S. New Insights for Drug Design from the X-Ray Crystallographic Structures of G-Protein-Coupled Receptors. Mol. Pharmacol. 2012, 82, 361–371.
(2) Dror, R. O.; Dirks, R. M.; Grossman, J. P.; Xu, H.; Shaw, D. E. Biomolecular Simulation: A Computational Microscope for Molecular Biology. Annu. Rev. Biophys. 2012, 41, 429–452.
(3) Buch, I.; Giorgino, T.; De Fabritiis, G. Complete Reconstruction of an Enzyme-Inhibitor Binding Process by Molecular Dynamics Simulations. Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 10184–10189.
(4) Sabbadin, D.; Moro, S. Supervised Molecular Dynamics (SuMD) as a Helpful Tool to Depict GPCR-Ligand Recognition Pathway in a Nanosecond Time Scale. J. Chem. Inf. Model. 2014.

gianniSupervised Molecular Dynamics (SuMD) as a helpful tool to depict GPCR-ligand recognition pathway in a nanosecond time scale
Read more

Performing molecular dynamics simulations using GPUs

Read the interview made to <a href=”https://es.linkedin.com/in/gdefabritiis”>Dr Gianni de Fabritiis</a>, group leader at the Computational Biophysics Laboratory of the Research Programme on Biomedical Informatics within the Barcelona Biomedical Research Park. He makes a nice overview of the things his group is doing and the tools he has developed, about volunteer distributed computing and its importance in research. He also highlights the potential importance of the use of GPUs in computer aided drug design and some of the challenges of the molecular dynamics simulation research field.

gianniPerforming molecular dynamics simulations using GPUs
Read more

Workshop in High-Throughput Molecular Dynamics Nov 7-8

During Nov 7-8th Acellera will host in Barcelona the first technical meeting in high-throughput molecular dynamics.  This event is sponsored by Acellera, NVIDIA and PRBB.  The technical meeting will go over a wide range of topics such as the use of Markov State models in MD simulations, the exploitation of parallel tempered metadynamics, or the design and use of plugins for ACEMD.  We look forward to meeting all of the attendees and to sharing recordings of some of the seminars with everyone else.    For more information visit here.

gianniWorkshop in High-Throughput Molecular Dynamics Nov 7-8
Read more

Nvidia GPU Test Drive VAR Registration

[breadcrumbs]

If you are an Nvidia GPU Test Drive Value-Added Reseller, please register
here to obtain a copy of ACEMD for your Testdrive system.

[php]
$mode=”testdrive”;
include( ‘/var/acemd/scripts/register/demo_request.php’);
[/php]

mattNvidia GPU Test Drive VAR Registration
Read more

Acellera at 2013 Biophysical Society Meeting in Philadelphia

At the meeting we will be presenting a poster showcasing our MD based platform for fragment based drug discovery. This platform is unique in that it can assign a ligand’s binding site (including selectivity of binding and specificity of the intereaction), and pose (of binding and any metastable states). Furthermore, for convergent systems, this platform can yield accurate estimates of the binding kinetics and thermodynamics. All of this is enabled by a combination of the simulation speeds that Metrocubo offers, and the skills we are developing in data analysis. To the best of our knowledge, no other platform can give such complete data sets.

For a sample video see:

For more information contact us at info@acellera.com

gianniAcellera at 2013 Biophysical Society Meeting in Philadelphia
Read more

Crucial step in AIDS virus maturation simulated for first time

Source: IMIM

Bioinformaticians at IMIM (Hospital del Mar Medical Research Institute) and UPF (Pompeu Fabra University) have used molecular simulation techniques to explain a specific step in the maturation of the HIV virions, i.e., how newly formed inert virus particles become infectious, which is essential in understanding how the virus replicates. These results, which have been published in the latest edition of PNAS, could be crucial to the design of future antiretrovirals.

HIV virions mature and become infectious as a result of the action of a protein called HIV protease. This protein acts like a pair of scissors, cutting the long chain of connected proteins that form HIV into individual proteins that will form the infectious structure of new virions. According to the researchers of the IMIM-UPF computational biophysics group, “One of the most intriguing aspects of the whole HIV maturation process is how free HIV protease, i.e. the ‘scissors protein,’ appears for the first time, since it is also initially part of the long poly-protein chains that make up new HIV virions.

Using ACEMD a software for molecular simulations and a technology known as GPUGRID.net, Gianni De Fabritiis’ group has demonstrated that the first “scissors proteins” can cut themselves out from within the middle of these poly-protein chains. They do this by binding one of their connected ends (the N-terminus) to their own active site and then cutting the chemical bond that connects them to the rest of the chain. This is the initial step of the whole HIV maturation process. If the HIV protease can be stopped during the maturation process, it will prevent viral particles, or virions, from reaching maturity and, therefore, from becoming infectious.

This work was performed using GPUGRID.net, a voluntary distributed computing platform that harnesses the processing power of thousands of NVIDIA GPU accelerators from household computers made available by the public for research purposes. It’s akin to accessing a virtual supercomputer. One of the benefits of GPU acceleration is that it provides computing power that is around 10 times higher than that generated by computers based on CPUs alone. It reduces research costs accordingly by providing a level computational power that previously was only available on dedicated, multi-million dollar supercomputers.

Researchers use this computing power to process large numbers of data and generate highly complex molecular simulations. In this specific case, thousands of computer simulations have been carried out, each for hundreds of nanoseconds (billionths of a second) for a total of almost a millisecond.

According to researchers, this discovery in the HIV maturation process provides an alternative approach in the design of future pharmaceutical products based on the use of these new molecular mechanisms. For now, this work provides a greater understanding of a crucial step in the life cycle of HIV, a virus that directly attacks and weakens the human immune system, making it vulnerable to a wide range of infections, and which affects millions of people around the world.

Reference:

Kinetic characterization of the critical step in HIV-1 protease maturation”. S Kashif Sadiq, Frank Noe and Gianni De Fabritiis. PNAS. DOI:10.1073/pnas.1210983109.

ACEMD Simulation details.

alejandroCrucial step in AIDS virus maturation simulated for first time
Read more

Penguin Computing Showcases Acellera’s ACEMD at SC2012

PENGUIN COMPUTING SHOWCASES ACELLERA’S ACEMD AT SC2012 LONDON, UK – 10 November 2012

Acellera today announced that Penguin Computing will showcase Acellera’s ACEMD on POD (Penguin Computing ON Demand) at Supercomputing 2012 in Salt Lake City, Utah, booth 3866. ACEMD is the world’s fastest and most efficient bio-molecular dynamics (MD) engine for GPU nodes and workstations. Specifically optimized to run on NVIDIA graphics cards, ACEMD enables on a single GPU performance equivalent to more than 100 CPUs while maintaining microsecond trajectories. In combination with POD, ACEMD aims to facilitate the transition of MD from a multi-node, single-trajectory science to high-throughput MD (HT-MD), a new paradigm in computational biophysics.

Matt Jacobs, Sr. Vice President of Corporate Development for Penguin Computing states, “As we approach the clock cycle limits on traditional CPU architectures, multicore and GPU technologies are moving to the forefront to reshape the way our customers conceptualize their research. Acellera’s ACEMD package demonstrates the true capabilities that this new level of parallelism offers and Penguin Computing is pleased to be able to showcase this package on our POD environment at this year’s Supercomputing conference.”

Molecular Dynamics simulation is a powerful technique for investigating biological systems with atomic resolution. Prior to the introduction of Acellera’s ACEMD, the cost of bio-MD simulations for R&D was prohibitive, requiring hundreds of CPU cores for a single run on dedicated HPC clusters. ACEMD transformed the bio-MD field by introducing highly optimized bio-MD simulation on comparatively low-cost GPUs, enabling supercomputing class simulations on a small GPU cluster. ACEMD enables new kinds of analyses, such as the complete reconstruction of protein–ligand interactions in terms of thermodynamics and kinetics, or MD based fragment-based drug discovery. Such studies open new avenues in pharmaceutical research, and illuminate ligand binding site and pose, as well as specific kinetic signatures that can be associated with the quality of small molecules during drug development.

“We are pleased to have this opportunity to work with such a reputable provider of world-class HPC solutions. Penguin’s POD technology combined with Acellera’s ACEMD provides an unbeatable platform for on-demand, highly-efficiency bio-MD simulation for all scientists, from those just beginning to explore its huge potential to established MD users” states Gianni De Fabritiis, Founder and Chief Science Officer at Acellera.

ACEMD is the engine behind GPUGRID, one of the largest distributed computing projects worldwide. ACEMD is stable, scalable, robust, and can handle thousands of highly demanding MD simulations daily. ACEMD simulations have helped researchers push our understanding of protein-ligand interactions, ion channel dynamics, membrane protein behavior, and mechanisms for drug resistance.

Penguin Computing and Acellera are looking to expand their partnership to include turn-key solutions which integrate Acellera and Penguin technology. This would mark Penguin Computing as the first U.S. Value Added Reseller and provider of L1 system support for Metrocubo, Acellera’s MD appliance.

To experience a demonstration of ACEMD at SC2012, visit Penguin Computing booth 1217 November 12-15th where ACEMD simulations will use POD, an HPC cloud that provides instant availability on-demand. Interested parties will be offered the opportunity during those same dates to challenge ACEMD with their own systems on POD. Contact David Soriano at info@acellera.com for more information.

About Penguin Computing
For well over a decade Penguin Computing has been dedicated to delivering complete, integrated High Performance Computing (HPC) solutions that are innovative, cost effective and easy to use. Penguin offers a complete end-to-end portfolio of products and solutions ranging from Linux servers and workstations to integrated, turn-key HPC clusters and cluster management software. For users that want to use supercomputing capabilities on-demand and pay as they go, Penguin offers ‘Penguin Computing on Demand’ (POD), a public HPC cloud that is available instantly and as needed. With its broad portfolio of solutions Penguin is the one-stop shop for HPC and enterprise customers and counts some of the world’s most demanding HPC users as its customers, including Caterpillar, Life Technologies, Dolby, Lockheed Martin, the U.S. Air Force, and the U.S. Navy.
To learn more, visit http://www.penguincomputing.com

About Acellera
Acellera is a UK based company focused on providing new technologies in the field of research and development. Since 2006, Acellera has innovated in the field of molecular dynamics simulation software for accelerator processors. Its current flagship product, ACEMD, delivers cluster-computer levels of performance for MD simulations on personal GPU workstations. Computational throughput is also optimized through the joined development of software and hardware. Recently, Acellera designed and patented a new enclosure for GPU optimized calculations for these purposes.
One mission of Acellera is to develop high-throughput molecular dynamics techniques that deliver solutions for estimating common physical chemistry properties as binding affinities, kinetics, poses and pathways with experimental accuracy. The development of new methods for molecular data analysis allows us to understand binding processes to a new level of insight. In this field, the Binding Assay solution by Acellera provides a level of details on the binding process that is unique worldwide.
Acellera is the owner and unique provider of the intellectual property behind all these technologies and the software, hardware infrastructure.
To learn more, visit http://www.acellera.com

Penguin Computing is a registered trademark of Penguin Computing, Inc. Penguin Computing on Demand is a pending trademark in the U.S. ACEMD is a trademark of Acellera. All other trademarks are property of their respective owners. Other product or company names mentioned may be trademarks or trade names of their respective companies.

alejandroPenguin Computing Showcases Acellera’s ACEMD at SC2012
Read more