Folding@home: Achievements from over 20 years of citizen science herald the exascale era

Abstract

Simulations of biomolecules have enormous potential to inform our understanding of biology but require extremely demanding calculations. For over 20 years, the Folding@home distributed computing project has pioneered a massively parallel approach to biomolecular simulation, harnessing the resources of citizen scientists across the globe. Here, we summarize the scientific and technical advances this perspective has enabled. As the project's name implies, the early years of Folding@home focused on driving advances in our understanding of protein folding by developing statistical methods for capturing long-timescale processes and facilitating insight into complex dynamical processes. Success laid a foundation for broadening the scope of Folding@home to address other functionally relevant conformational changes, such as receptor signaling, enzyme dynamics, and ligand binding. Continued algorithmic advances, hardware developments such as graphics processing unit (GPU)-based computing, and the growing scale of Folding@home have enabled the project to focus on new areas where massively parallel sampling can be impactful. While previous work sought to expand toward larger proteins with slower conformational changes, new work focuses on large-scale comparative studies of different protein sequences and chemical compounds to better understand biology and inform the development of small-molecule drugs. Progress on these fronts enabled the community to pivot quickly in response to the COVID-19 pandemic, expanding to become the world's first exascale computer and deploying this massive resource to provide insight into the inner workings of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus and aid the development of new antivirals. This success provides a glimpse of what is to come as exascale supercomputers come online and as Folding@home continues its work.

Figure 1

Running M simulations of length t₁ can provide as much information as running one simulation of length M × t₁, providing an enormous speedup if one runs parallel simulations over a single long simulation. (A) A simple three-state landscape where $k_{12}$ is the rate of transitioning from state 1 to state 2 and $k_{23}$ is the rate of transitioning from state 2 to state 3. (B) Speedup versus number of simulations for this simple landscape. The key conclusion is that the more two state a system is (i.e., the more one barrier is much larger than the other), the closer one comes to linear scaling, where running M simulations of length t₁ is equivalent to running one simulation of length M × t₁. Shown are plots for a range of $r=k_{12}{/k}_{23}$, where $k_{12}$ is the rate constant of the slow barrier crossing, and $k_{23}$ is the rate constant of the fast barrier crossing. Reproduced from Shirts and Pande (11).

Figure 2

The 10 highest-flux folding pathways for the 39-residue protein NTL9, which folds on a ms timescale. State sizes are proportional to the population, arrow widths are proportional to the flux, the colored ribbon shows a representative structure, and the gray structures convey the extent of structural diversity. Reproduced from Voelz et al. (3). To see this figure in color, go online.

Figure 3

The 10 highest flux pathways from the unbound states of the lysine-, arginine-, and ornithine (LAO)-binding protein to the arginine-bound state. The arrow widths are proportional to the flux, the two lobes of the lysine-, arginine-, and ornithine-binding protein are shown in dark/light blue, and arginine is shown as red sticks. State numbers and their equilibrium populations are also shown. The conformational selection and induced fit pathways from the unbound states to the encounter complex state are shown in green and gray arrows, respectively. Reproduced from Silva et al. (32). To see this figure in color, go online.

Figure 4

A structure with a cryptic pocket in Ebola’s VP35 protein (blue) overlaid with the crystal structure (gray). The existence of this pocket and its allosteric control over RNA binding were confirmed using chemical labeling experiments. Reproduced from Cruz et al. (92). To see this figure in color, go online.

Figure 5

Structural states with the Nsp16 cryptic pocket closed and open, showing how pocket opening is correlated with collapse of the active site’s S-adenosyl-L-methionine (SAM)-binding pocket. The insets show surface views of the closed and open pockets. Residues exposed upon pocket opening are shown in cyan, and the regions undergoing the opening motion are shown in blue. SAM is in magenta sticks, and the RNA substrate is in green sticks. Collapse of the SAM-binding pocket is measured as the distance between two loops labeled SAMBL2 and gate loop 2, shown in yellow. Reproduced from Vithani et al. (99). To see this figure in color, go online.

Figure 6

The probability distribution of spike opening for three spike homologs. Opening is quantified in terms of how far the center of mass of a receptor-binding domain deviates from its position in the closed (or down) state. The cryptic epitope for the antibody CR3022 (red) and the ACE2-binding interface are both exposed in open structures but buried in closed structures. As a result, more open spikes are better at cell entry but more susceptible to host immunity. Reproduced from Zimmerman et al. (2). To see this figure in color, go online.