Folding@home

New paper #61: Computational screen to identify important mutations in influenza

We're happy to announce a new Pande Group paper (paper #61 at http://folding.stanford.edu/English/Papers).  This paper describes a new computational screen to identify important mutations in influenza:

Combining Mutual Information with Structural Analysis to Screen for Functionally Important Residues in Influenza Hemagglutinin. 
Peter M. Kasson and Vijay S. Pande. Pacific Symposium on Biocomputing 14:492-503(2009). 
Download URL: http://psb.stanford.edu/psb-online/proceedings/psb09/kasson.pdf  

The influenza hemagglutinin protein performs several important functions, including attaching the virus to cells it will infect and releasing the viral genome into the interior of the cell. Most protective antibodies against influenza also bind to the hemagglutinin protein. We wish to understand how mutations to hemagglutinin affect viral function, including what keeps avian influenza ("bird flu") from being readily transmissible between humans. In this paper, we have applied a technique from information theory known as mutual information to genetic sequence data to predict important mutation sites on the hemagglutinin protein. In follow-up work, we are combining this technique with other methods to refine these predictions and test some of them using Folding@Home. 

PS For those curious out in more details, check out the paper (see link above) or the technical abstract:

Influenza hemagglutinin mediates both cell-surface binding and cell entry by the virus. Mutations to hemagglutinin are thus critical in determining host species specificity and viral infectivity. Previous approaches have primarily considered point mutations and sequence conservation; here we develop a complementary approach using mutual information to examine concerted mutations. For hemagglutinin, several overlapping selective pressures can cause such concerted mutations, including the host immune response, ligand recognition and host specificity, and functional requirements for pH-induced activation and membrane fusion. Using sequence mutual information as a metric, we extracted clusters of concerted mutation sites and analyzed them in the context of crystallographic data. Comparison of influenza isolates from two subtypes—human H3N2 strains and human and avian H5N1 strains—yielded substantial differences in spatial localization of the clustered residues. We hypothesize that the clusters on the globular head of H3N2 hemagglutinin may relate to antibody recognition (as many protective antibodies are known to bind in that region), while the clusters in common to H3N2 and H5N1 hemagglutinin may indicate shared functional roles. We propose that these shared sites may be particularly fruitful for mutagenesis studies in understanding the infectivity of this common human pathogen. The combination of sequence mutual information and structural analysis thus helps generate novel functional hypotheses that would not be apparent via either method alone.

Visit from Jim Clark and John Hennessy (President of Stanford University)

Jim Clark has been a major donor to Stanford and his great contributions has had a huge impact on my group's work in general and on Folding@home in particular.  Jim Clark was a professor of CS at Stanford, but subsequently was involved in many successful major Silicon Valley companies (see his wikipedia page for all the details).  He also donated over $100M to Stanford to build the Clark Center, a University-wide center for interdisciplinary biology.  

Today, Dr. Clark along with John Hennessy (the President of Stanford University) visited our offices to hear about our recent work.  They both were heavily involved in computer architecture in the past, so they were interested to hear about our work with GPUs and the success we are seeing there (in particular, the significant speed increases).  Also, they are both interested in neuroscience and so I was excited to tell them about our recent Alzheimer's work.

Anyway, I was excited to give them both an update and some idea of where we're going, and it was great fun for me to tell them all about how much we've done.

New paper #60: PS3 paper

We have another paper that came out recently (paper #60 at http://folding.stanford.edu/English/Papers).  This one describes the PS3 client in FAH:

Accelerating Molecular Dynamic Simulation on the Cell processor and PlayStation 3 
by Edgar Luttmann, Daniel L. Ensign, Vishal Vaidyanathan, Mike Houston, Noam Rimon, Jeppe Øland, Guha Jayachandran, Mark Friedrichs, Vijay S. Pande. 

'In this paper, we detail how we were able to get great speed increases for Folding@home (and actually certain molecular dynamics calculations in general) on the PS3. This is our first paper using the PS3, laying out the "how does it work," with a follow up paper in the works describing the results obtained in FAH from PS3 clients. 

It is also worth noting that this paper is a collaboration between FAH team members (Luttmann, Ensign, Vaidyanathan, Houston [now at AMD], Jayachandran, Friedrichs, and Pande) with developers at Sony (Rimon and Øland and their coworkers). 

Here's the more technical abstract as well:

Implementation of molecular dynamics (MD) calculations on novel architectures will vastly increase its power to calculate the physical properties of complex systems. Herein, we detail algorithmic advances developed to accelerate MD simulations on the Cell processor, a commodity processor found in PlayStation 3 (PS3). In particular, we discuss issues regarding memory access versus computation and the types of calculations which are best suited for streaming processors such as the Cell, focusing on implicit solvation models. We conclude with a comparison of improved performance on the PS3's Cell processor over more traditional processors.

Folding@home Alzheimer's Simulation work published

I am very happy to announce that a first key work (paper #58 at http://folding.stanford.edu/English/Papers) from the Folding@home project on Alzheimer's Disease (AD) was just published

Simulating oligomerization at experimental concentrations and long timescales: A Markov state model approach, by Nicholas W. Kelley, V. Vishal, Grant A. Krafft, and Vijay S. Pande.   J. Chem. Phys. 129, 214707 (2008); DOI:10.1063/1.3010881 

url:  http://link.aip.org/link/?JCP/129/214707

Abeta misfolding and aggregation is believed to be the cause of Alzheimer's Disease. Simulations, like Folding@home, are a natural way to understand this process. However, there are several key challenges for simulating the key step -- oligomerization. This work represents a new way to simulate Abeta oligomerization, with a key advance of being able to simulate experimentally relevant timescales and concentrations, using a novel method. We use this new method and the power provided by Folding@home donors to simulate oligomerization in all-atom detail. This has lead to specific predictions about the process, which we are now testing experimentally. 

In many ways, this paper is the "tip of the iceberg" for the Folding@home activities in AD, with a lot more interesting results to come, especially in terms of experimental tests of our predictions and interesting new possibilities for new drugs and AD therapeutics.  So, while we're excited that this result is now past peer review, we're even more excited for what's coming down the pipeline, waiting peer review.  We'll keep you posted as more results become public, hopefully with some even bigger announcements in 2009.

UPDATE:

It was asked which clients participated.  This work started several years ago and took some time to analyze and then publish.  So, it ran exclusively on classic clients.  For the follow up simulations, we are using a mixture of GPU, SMP, and classic clients.  Due to the large number of classic clients, they allow us to calculations not possible on the other platforms.  However, the raw speed (but smaller number) of the GPU and SMP clients allow us to get a good rough idea quickly, refining later with classic clients.

Long term vision for FAH: Science, clients, etc

It's been a while since I've posted on this topic, so I thought it would be a good time to talk about the long term vision for FAH.  In the first few years, we have done a lot to build up methodology (novel distributed computing algorithms, tests of methods), especially in the last years working to bring GPU's, PS3's, and tightly coupled SMP calculations to a distributed computing platform.  While all of this has been going on, there has been a lot of work which would have a direct impact to disease and human health, and donors should see more results (i.e. papers) coming out in this direction soon.  In particular, results on Alzheimer's Disease, Huntington's Disease, and some other surprises in the works (sorry, some details are best left until peer review is done).

In terms of a long term vision for the FAH software, our short term goals is to shore up the SMP client.  Just as there was a big improvement from the GPU1 -> GPU2 client, we have been working steadily on an SMP2 client -- a version which is much easier to use, requires less donor effort, and scales much better.  This really is a combination of a modified client and (especially) new cores.  This effort has been going on for about a year and it's far enough along that I'm starting to talk about it publicly.

In the previous post, I also mentioned about some of our work behind the scenes in revamping the backened of FAH, i.e. a whole new set of server codes.  We have code for the workserver and collection servers, but in time we expect the full backend to be cleanly rewritten from scratch.

Longer term, with a more solid GPU2 and SMP2 clients done, we can then move on to other areas.  Our goal is to be bleeding edge in the nature of our science, but get back to the simplicity and stability found in the classic client.  We still have a ways to go, but I am excited that projects that were started some time ago are now looking like they will see the light of day.

GPU2 WU's in "production mode"

As of last Monday, the GPU2 projects (both NVIDIA and ATI) have been in production mode, which means that we've moved on from testing to direct research on them.  The ATI cards have bee in production for a while (more than one month before) and we're starting to write our first peer reviewed scientific paper deriving from the results.

We're very excited about the GPU2 possibilities right now, since we are getting incredible production from GPU2 clients.  Once the first paper has been formally accepted by peer review, I'll talk more about the results.

Sharing FAH results -- raw data now available

We've been working behind the scenes for a while to find the right way to share the raw data results from Folding@home.  We've partnered with the Simbios National Biomedical Computing Center to provide data for download.  If you're curious, check out our first data set project page

https://simtk.org/home/foldvillin

We'll be releasing more data as time goes on.  Our hope is that by making the raw data openly available, this will greatly supplement the published results.

FAH/SMP Q & A

There was a good question in the forum that I thought others would be curious to hear:

From Vijay's blog entries it would seem that the SMP client has some fundamental advantages over running multiple singlecore client, but I can't really think of how that might be. Do you know of some architectural overview of how the MPI stuff is being used in this context?

We could just run multiple independent clients, but this would be throwing away a lot of power.  What makes an SMP machine special is that it is more than just the sum of the individual parts (CPU cores), since those cores can talk to each other very fast.  In FAH, machines talk to each other when they return WUs to a central server, say once a day.  On SMP this happens once a millisecond or so (or faster).   That 86,000,000x speed up in communication can be very useful, even if there isn't 100% utilization in the cores themselves. 

The easy route would have been to run multiple single-CPU FAH-cores (this is what other projects do), but that would be a big loss for the science, as this throws away a very, very powerful resource (fast interconnects between CPUs).  Indeed, it is this sort of fast interconnect which makes a supercomputer "super", since the CPUs in supercomputers (eg BlueGene) are pretty slow, but the communication between cores is very, very fast. 

We've done a lot to develop algorithms for FAH-style internet connections between CPUs, but there are some calculations which require fast interconnects, and that's where the FAH/SMP client is particularly important.  By allowing us to do calculations that we couldn't do otherwise, the science is pushed forward significantly (and we thus reward SMP donors with a points bonus due to this extra science done and the extra hassle involved in running the SMP client).

I guess it remains to be seen if we can pull off MPI on FAH to the point where it works effortlessly, but so far Lin and OSX look pretty good, so we're close.  The A2 core should hopefully seal the deal.  Now, the main task is getting Windows/SMP behaving well ...

Stanford Pervasive Parallelism Lab (PPL)

On Friday, Stanford launched the Pervasive Parallelism Lab (PPL).  There's been lots of press describing it.  The general plan for the lab is to develop a common paradigm for programming new architectures like GPU's, the Cell processor, Intel's Larabee, as well as multi-core CPUs.  This is something we at FAH are very interested in, as we have had to have a unique code path for each of these (i.e. a separate code for the high performance part of the ATI GPU, NVIDIA GPU, PS3, and SMP).  Having a single code path for all would be very, very exciting to keep FAH code development onto new hardware going smoothly.

Folding@home and Simbios

We've teamed up with Simbios (an NIH Roadmap center at Stanford) to start making key technologies in Folding@home available for others to use.  Our protein folding work has been named one of Simbios' key Driving Biology Problems (DBPs) and there's now a protein folding DBP page.  Our hope is that our developments which have been so useful in Folding@home can help accelerate the work of others as well.  Right now, there are two key areas we've been working to distribute, with more coming along the way.

First, we're working to make our GPU code available to others.  This code will be distributed in a couple of forms.  First, we'll give out a GPU-enabled version of Gromacs (basically, a standalone version of the GPU2 core), which will enable others to get major speed increases from GPUs.  Next, we are working to release a GPU-enabled library (OpenMM), which will allow others to integrate GPU code into their programs.  OpenMM is special in that it is a place for integrating both application developers as well as GPU vendors; much like OpenGL, our hope is that hardware acceleration vendors will now have a single API to accelerate and people who want to write applications will have a single, hardware accelerated API to use that would work on a variety of platforms.

Second, we're also starting to make large data sets from Folding@home available to others.  You can see some of the first data sets on this project page, and we expect to put more data up as time goes on.  Folding@home donors have generated wonderful data sets that aren't possible to generate by other means, and our hope is to publish them so that other scientists can data mine them for other purposes as well.

Finally, we expect to release more technologies from FAH.  In particular, we will be releasing some of the key server-side algorithms which allow FAH to use lots of processors to tackle single complex problems.  This will allow people to run massively parallel code on large clusters to do tasks similar to FAH (albeit on a smaller scale).