Folding@home

How FAH works: Thermodynamic alchemy

One of the remarkable strengths of thermodynamics is that it is in many ways more general than many of the laws of physics.  One of the key concepts in thermodynamics is that of a state function.  The key aspect of a state function is that if you're interested in the differences in a state function between some initial and final state, it doesn't matter what you do to get from the initial to the final -- whatever path you take would get you the same value for the difference of the state function.

How is this useful?  Well, often we can take paths in simulations which are not possible in reality, but since the values of state functions only depend on the initial and final state, as long as those as the same as in reality, we get the same result!  One key way we use to take advantage of this is thermodynamic alchemy.  For example, often one is interested in the difference between an amino acid could be a single methyl group.  In simulation, we can turn a proton alchemically into a methyl group and calculate everything we need thermodynamically, even if this isn't possible to do experimentally in a direct way!

What can Thermodynamics tell us?

Thermodynamics (and it's "older brother" statistical mechanics -- but I'll refer to both as Thermodynamics) plays a very important role in what we do.  I thought I'd talk a little about it.  Thermodynamics plays a major role in several aspects of work.  What thermodynamics helps us answer is what the eventual probability of finding various states would be in equilibrium (i.e. if we wait long enough for everything to settle down).  For example, what is the probability of finding the folded state? 

First, let me talk a bit about how to think of this concept of probability.  One way to think of this is in terms of ensembles -- if there's a 70% chance of finding the native state, that means that if we ran 100 simulations and looked at them to see which ones folded, 70 of them would have folded.  We can also think of this in terms of time -- if there's a 70% chance of finding the native state, that means that if we look at a single protein, 70% of the time we'll see it in the folded state.  Either way works and often we use these two ideas interchangeably. 

WIth knowing about thermodynamics, we can develop and apply new tricks which let us run our calculations much more efficiently.  In particular, sometimes the probabilities we're looking for is 1 in a million.  In cases like that, it's not efficient at all to run millions of simulations to look for a few rare cases.  Instead, with a knowledge of thermodynamics, we have methods that allow one to use many fewer simulations and still find these rare events.

I teach Thermodynamics to juniors at Stanford in the fall and Statistical Mechanics to graduate students in the Spring.  The nature of the course is the basis behind the research that we do.

Nanomedicine center

Our lab is part of an NIH roadmap center for Nanomedicine.  The basic idea of the nanomedicine roadmap centers is to use nanotechnology and other advances in chemistry, structural biology, and cell biology to make advances in medical applications.  The center's focus is protein folding, and in particular protein folding in vivo, i.e. in the presence of chaperones (see my post a few days ago talking about Del Lucent's work with chaperones in FAH).  One can learn more about the center here.

We're meeting today to present our recent results relevant for protein folding in the nanomed center and to make future plans.  FAH has been able to make significant advances in our understanding of how chaperones may work to speed folding, and we're looking forward to the future where we apply these ideas to specific biomedical applications.  In particular, we have a project which we think could (if sucessful) lead to an interesting new drug for cancer.  The way this works is that cancer cells need chaperones really badly, as they are growing so quickly.  If one can inhibit chaperones, one can kill cancer cells (hopefully without doing too much damage to healthy cells).  This approach has been used before, but we hope to use our unique methods to make a major advance in this area.

PS here's the mission of the nanomed center.  I'm reproducing it here to give some better idea of the general goals of the center and to give some idea of what we're collectively trying to do.

The chaperonin TRiC is a protein folding nanomachine necessary for the growth of all eukaryotic cells. The protein substrates of this barrel-shaped 16-subunit nanomachine include actins, tubulins, and tumor suppressor proteins. Similarly to its archaeal homolog Mm-cpn, it couples ATP hydrolysis to internalization, folding, and release of newly synthesized polypeptide chains. The folding cycle includes opening and closing of a built-in lid of the chaperonin critical for binding and release.

Our Nanomedicine Development Center (NDC) will extend and integrate the current techniques in electron cryomicroscopy, single-molecule imaging, computational biology, and X-ray crystallography to quantify the chaperonin subunit conformations and dynamics as well as the protein folding intermediates bound within the chaperonin cavity. The marriage of these advanced technologies will allow us to visualize chaperonin machinery functioning not only in vitro but also within cells.

Building on a more comprehensive and quantitative description of these protein folding nanomachines, we will engineer modified chaperonins to provide a novel therapeutic tool for inhibiting and promoting the folding of selected proteins whose misfolding or aggregation are associated with human diseases. These proteins include actin, tumor suppressor proteins p53 and Von Hippel Lindau, the aggregating A-beta peptide and the cataract related lens protein, γ-crystallin.

Through visualizing conformations and sites in which these chains are folded by the chaperonin, together with their experimentally observed and computed dynamics, we will also design novel substrates that will be folded efficiently in the naturally occurring or newly engineered chaperonin, opening new avenues in protein design. The corresponding approaches will include design of adaptor peptides for modifying the substrates or chaperonins to enhance or inhibit substrate-chaperone interactions.

We have assembled a team of 15 investigators from 6 institutions with expertise in chaperones, protein folding, electron cryomicroscopy, computer simulation and modeling, X-ray crystallography, single-molecule imaging and trapping, and clinical research. We will work together in developing a strategic set of experimental and computational tools that will enable characterization of biological nanomachines, both in vitro and in vivo. We will interact with complementary expertise of other NDC in synthetic chemistry and fluorescence technologies.

Our Center can be a critical resource to other NDC who need assistance in solving protein folding and aggregation problems. Our clinical investigators will contribute to the design of pilot studies for therapeutic applications in cell culture models of disease states. We also plan new educational tools via virtual courses in design and application of biological nanomachines, aimed at bridging the gap between biology and mechanical engineering for students in our 6 participating institutions. Finally, the organization, collaborations and communications in our NDC exemplify the 21st century goal of conducting interdisciplinary research via new mechanisms of data sharing and analysis.

For more information, go to the center's web site at http://proteinfoldingcenter.org/

v6 making good progress

We're finally set to start rolling out version 6 (v6) of Folding@home.  You can download a beta version from our download page.  There are several changes to v6, many under the hood, and some that are more visible.  We'll be talking more about v6 as we roll it out further, but I wanted to make a few comments here.  First, we're working to do better integration of SMP to the regular client and in particular to make the SMP client more flexible.  The SMP client/core is very much the future of FAH and so a lot of work has been going into it.

Our hope is that our beta testing for v6 will go smoothly.  I'll be writing here about specific v6 features as time goes on.

How FAH works: Molecular dynamics

Molecular dynamics (MD) is a major part of FAH.  Most (but not all) of the FAH cores are MD engines.  The basic idea behind MD is to simulate reality by following how atoms would move around, driven by Newton's equations.  This is much like trying to predict how a pool game would go by simulating how the balls move around the table.  This only works if you know what are the forces being put on the balls and this is the same in MD -- we need to know the forces that the atoms exert on each other.

One of the key assets of MD is that it can predict kinetics, i.e. not just what the final outcome would be (thermodynamics) but how it happened.  This is particularly important for studying protein folding and protein misfolding, as this is precisely what we're interested in.  However, there are other cases where we care purely about the end in a statistical manner (eg out of 1,000,000 drug molecules, what fraction would bind to this protein?).  While there are other methods that could be used in this case (eg Monte Carlo methods), MD can often be more efficient than Monte Carlo (MD has a better "move set" than MC and MD codes tend to be better optimized and computationally efficient).

There are lots of other issues to deal with and MD Is just one of them.  I'll talk about other issues in the future.  In particular, it's important to ask where do the forces come from that we use.  This brings up the question of force fields, which I'll leave for a post in the future.

Meet FAH team member Dan Ensign

Dan Ensign joined the group two years ago and has quickly become a key member of the Folding@home team.  Dan's work centers around one of the core scientific areas in the group: understanding how proteins fold.  Dan is using all possible tools at his disposal, including regular FAH clients, SMP clients, GPU clients, and PS3 clients.  Dan's first paper using SMP clients just got accepted for publication and I'll highlight it in a little more detail below (but it really deserves its own post, which I'll leave for another date).  He is now turning a lot of attention to the analysis of all the piles and piles of data for the PS3 client and especially to work to continue to improve the calculations that are done on the PS3.  Indeed, his results were key in our scneitific improvements from v1.1 to v1.2 (current version).

Dan's work with SMP clients has lead to a phenomenal data sets for protein folding, including trajectories that are both large in number and very long.  This data set (which most people may have thought would be impossible to simulate) will play a key role in both understanding how small proteins fold, as well as in developing methods to further push FAH to study more and more complex proteins (as well as related questions such as misfolding and lipid vesicle fusion) by making more enhanced MSM methods.

Dan's work is pretty wide spread so I will likely make some more posts about it in the future.  For now, I'll leave you with a link to a youtube video where Dan's discussing some of the science behind FAH.

How FAH works: Folding on streaming processors (GPU's and PS3)

One of the critical issues in computer science right now is the limits to how fast a single CPU can calculate.  While Moore's law is still going strong in a literal sense -- i.e. the number of transistors which one can put on a chip is doubling every 2 years or so -- this doubling of transistors is not leading to a doubling in CPU speed as it used to over the last few decades.  Well, at least not for typical programs (eg Microsoft Word).  In order to get big speed increases, there'sa major change in the programming paradigm.   One key change is the existence of "streaming processors."  GPU's and the Cell processor in the PS3 are both examples of streaming processors.

What makes streaming processors potentially much faster than regular CPU's is how they handle computation vs memory access.  Normal CPU's use lots and lots and lots of transistors on cache (local memory on the CPU chip to help keep the CPU fed with data and instructions).  Streaming processors use the additional transistors on additional computing elements (eg floating point units).  By doing so, they can do lots of FLoating point OPerations per second (FLOPS) in an optimal situation, although getting one's code to behave optimally is not easy.  Typically this means balancing FLOPS with memory access to make sure that there's data available for calculation.  This has been the primary challenge in our GPU and PS3 codes, and is something which we have, for the most part, figured out for a significant subset of the calculations we run on FAH.

These advances have lead to our GPU and PS3 clients.  The family history of all of this starts with the GPU core.  This GPU code was then brought over to the PS3 and enhanced.  We are working to bring back some of those scientific enhancements back to the GPU code.  This is all pretty bleeding edge, but so was distributed computing in 2000 when we started.  Our expectation is that given how modern processors are developing, in 8-10 years streaming processors will be much more standard and will be a major way in which FAH works. 

Finally, it's also interesting to think of how CPU's may themselves turn into streaming processors.  As CPU's add more cores, they start to have more functionality similar to streaming processors in a limited way.  Perhaps more interesting are some of the new chips rumored to be developed at Intel and AMD/ATI.  Intel's 80-core chip is very interesting and something which our code would likely run well on.  Also, the fusion of AMD's CPU's with ATI's GPU's could be very exciting, potentially bringing the best of both worlds.  We're looking forward to these and lots of other emerging technology.  FAH is running very fast now (over a petaflop, i.e. 1,000,000,000,000,000 floating point operations per second !) and we look forward to continuing to push the frontiers.

Meet FAH team member Dr. Peter Kasson

Dr. Peter Kasson, MD, PhD has been working in the group for almost two years.  Peter's interests are in the area of lipid vesicle fusion, a process relevant for many biological processes as well as relevant for disease and infection.  Lipid vesicles are large assemblies of detergent-like molecules that are used to house and/or contain many different types of molecules in biology.  Many viruses ("envelope viruses") are housed in lipid vesicles, but so are the neurotransmitters in our brains.  In order for these containers to be shuttled around (eg as neurotransmitters transmit thoughts in the brain or when viruses try to enter cells) lipid vesicles fuse with other vesicles or with cells (which are like giant lipid vesicles since cell walls are made of lipids).

Thus, membrane fusion provides the mechanism for the entry and infection by enveloped viruses such as influenza, Ebola, SARS, and HIV.  Not surprisingly, the steps driving membrane fusion are the targets by which these critically important processes are regulated, and thus a fundamental understanding of membrane fusion mechanisms will enable the design of novel anti-viral drugs. When faced with the potential for rapidly emerging and highly virulent infections, such an understanding of fusion becomes of particular importance to both fundamental biology and national health.

Peter brings a diverse set of skills to the group.  He has both a MD and a PhD in biophysics, which makes him particularly well suited to study lipid vesicle fusion from both a biophysical (eg model membrane systems) and a medical (eg looking at models for influenza infection) point of view.  Peter also plays a large role in the infrastructure development for Folding@home.  He is the lead developer for the SMP client & core and is constantly looking for ways to push FAH's capabilities in order to tackle these complex problems.  Indeed, due to the very large size of lipid vesicle simulations (which can easily be 1 to 10 million atoms in size), the SMP client is extremely important to this research.

Peter has already had several papers accepted in peer review with his work from FAH, examining the biophysical nature of lipid vesicle fusion as well as studying the protein-membrane interactions involved in the fusion caused by the influenza virus.  Finally, what many people may not realize is that a large aspect of lipid vesicle fusion in biology involves protein conformational change and/or some sort of folding.  Indeed, in influenza, there is a small peptide (HA) which likely goes through some folding process during fusion and subsequently this new HA peptide conformation leads to a change in the host membrane, making it more susceptible to fusion, leading to influenza infection, and thus getting the flu.
  Studying this process brings together many aspects of what we've learned in FAH and builds on top a lot of our previous work, especially in peptide folding and MSM creation.

How FAH works: Markov State Models

This post may get a little technical, but I wanted to start a new set of posts describing the inner workings of Folding@home.  To say FAH is complex is in many ways an understatement.  On the surface, FAH does seem a lot like other distributed computing projects: there's lots of WU's which go out to client machines and they get calculated and then come back.  However, there are a lot of differences going on in FAH.

One of the principal challenges in FAH is that we're trying to use lots of processors to in a sense speed up a calculation which many would have thought was intrinsically serial, i.e. that could only be done by a single very, very, very fast processor.  The reason why is that we are studying how proteins change in time, and it's hard to parallelize the 23rd step if you haven't completed the 22nd step, etc.

However, through the years we have been developing ways to solve this issue.  In the last few years, we have made significant progress in a method called "Markov State Models" or MSMs for short.  MSMs in a sense allow us to parallelize these seemingly intrinsically serial tasks.  The way this works is that we build a kinetic model for the process, dividing up the possible dynamics into a series of states (related protein conformations) and rates between these states.  The rates are what's calculated in FAH WU's.  Once we have all this data, we need to run some fairly sophisticated Bayesian Machine Learning methods to identify what are reasonable states and then to calculate the rates between them.

We have had a several recent advances in MSM methodology and those papers are on our papers website.  We have also had several MSM applications including studying protein folding, lipid vesicle fusion, and abeta aggregation (Alzhiemer's Disease simulations).  While we will continue to improve our MSM methodology, we are very much excited about the potential applications and have great thrusts going in both areas.

Meet FAH team member Del Lucent

Del Lucent has been with the group for 3 years and has been working to understand how proteins fold in vivo (i.e. how they fold in cells and how that might be different than folding in a test tube).  On the surface, this may not seem like something which is very amenable to simulation (it's easy to simulate a protein in isolation -- a good model for proteins in a test tube -- but really hard to simulate a whole cell).  Therefore, we must put not all the details but what we think are the right details. 

Del has been concentrating on studying how folding differs in the presence of chaperonins.  Chaperonins are pretty neat molecules (see the picture on the right).  First off, they're HUGE -- made up of multiple proteins assembled together into a big container.  Proteins go into the container misfolded and somehow come out folded.  That's what we want to understand -- how do chaperonins help proteins fold?  You can check out Del describing this in more detail on youtube.

Del's first paper in this direction was recently published in PNAS (see http://folding.stanford.edu/papers.html -- his is #50).  It shows some surprising results for simple models of chaperonins, but Del is now studying more detailed models.  Del is also a part of our team studying folding in the ribosome (another part of the cell which makes folding in vivo different).  Finally, Del is also interested in using his knowledge about chaperonins and folding to help develop new therapeutics for cancer (more specifically -- it turns out that chaperonins are heavily used by tumor cells as they are growing so quickly and coming up with ways to inhibit chaperonins has become an important new type of therapeutic to fight several forms of cancer).

Del also plays a major role in building Folding@home infrastructure, as he is the current "stats  master" and he works tirelessly with other members of the group to make sure that the stats are right, and to improve them as we improve our clients (eg lots of changes for v6).