How it works
What is Folding@home?
Folding@home is a distributed-computing project that simulates how proteins move. Volunteers all over the world install a small client on their computers; while their machines are otherwise idle, the client runs short pieces of molecular-dynamics simulation and reports the results back to our research servers. Stitched together, those pieces add up to one of the largest scientific computing systems in the world.
The motivation is biology. Proteins are the molecular machines that run living cells, and most of what a protein does depends on how it folds and how it moves. When folding goes wrong, you get disease — Alzheimer's, Huntington's, certain cancers. When folding works normally but the motion is hard to capture experimentally, you can miss the very feature a drug needs to target. Simulation gets at all of that.
What is protein folding?
A protein is a chain of amino acids. Newly built, it's a linear string, but it folds within seconds into a precise three-dimensional shape determined by the sequence. That shape is what lets it function — binding a substrate, recognizing a virus, catalyzing a chemical reaction. Misfold and the function is lost or, worse, becomes harmful: protein-misfolding diseases like Alzheimer's and Parkinson's are driven by proteins that take on the wrong shape and aggregate.
Folding happens on timescales of microseconds to milliseconds, with atoms moving every few femtoseconds — a billion-fold gap between the steps you have to take and the answer you want. That gap is why simulating folding is so computationally expensive, and why our approach (many short simulations in parallel, combined statistically) works where single long simulations would not. More on the science →
How is a global supercomputer is made of laptops?
Real-world proteins move on timescales millions of times longer than any single computer can simulate quickly. Folding@home gets there by running many short trajectories in parallel instead of one long one, then stitching them back together statistically.
- Project assigned. A research target is broken into Work Units — identified as
Project (Run, Clone, Generation)— each a short slice of the simulation. - Donor folds. Your client downloads a Work Unit, simulates a few nanoseconds of motion, and uploads the result.
- Trajectories ensemble. Thousands of donor results combine into a Markov State Model of the protein's full motion.
- Researcher learns. Lab teams query the model for druggable pockets, misfolding pathways, mutation effects — and publish.
Result: timescales that used to take a national lab months now take volunteers days.
Why not just use a supercomputer?
A modern supercomputer is essentially a cluster of CPUs connected by fast networking. The processors aren't dramatically faster than the one in your PC — supercomputers win on having a lot of them and on tight coupling between them. That tight coupling is great for problems where you need one giant simulation; it's actually wasted on F@h's approach, which needs many simulations and doesn't need them to talk to each other.
Protein-folding dynamics is statistical: a single long simulation would tell you one possible folding pathway, not the distribution of pathways the molecule actually samples. F@h gets the distribution by running thousands of independent short trajectories on volunteer hardware — computationally cheaper, scientifically richer, and accessible to any lab without a supercomputer allocation. More on why this works →
Who owns the results?
Folding@home is a nonprofit research project. We don't sell the data and we don't make money from it. Simulation results are analyzed and submitted to scientific journals; the resulting papers are listed on the papers & results page. After publication, raw simulation data is shared with other research groups on request, and the largest datasets are made publicly available. Key infrastructure (the MSMBuilder analysis library, the Copernicus workflow engine) is open source for anyone to use, including labs unaffiliated with F@h.