August 13, 2002 | To get a sense of the computational power required to perform protein-folding calculations, one needs to look in more detail at what exactly is being simulated. A typical protein can have 30,000 atoms, surrounded by perhaps 500,000 water molecules.
With Blue Gene, IBM takes a molecular dynamics approach to studying protein folding. Each simulation starts with a model of a protein molecule within a solvent. In the real world, such a protein molecule would start to fold (or change shape) based on chemical attraction between the discrete atoms in the protein and the solvent molecules. To simulate protein folding, the effects of these chemical interactions are calculated for every atom in the protein and solvent.
The simulation program must take into account the forces on each atom and calculate each atom's position after these forces are applied over a time step (a set period of time). This process is done for every atom in the protein under investigation.
After a time step is completed, new forces are calculated and these forces are then applied to each atom again. The entire process is repeated multiple times. The end result is a calculation of the trajectory of all the atoms, so researchers can see where each atom is located as a function of time.
One of the key factors in a simulation of this type is the length of a time step. The bottom line is that smaller time steps yield a more accurate simulation. That's because each atom is influenced by all of the atoms around it; every time neighboring atoms move, the forces on a specific atom change. But a smaller time step means many more calculations must be done to simulate how a protein will ultimately behave over time.
So scientists must find some compromise. They must select a time step that is short enough to accurately account for the dynamics of the atoms in the protein, balanced against the number of computations that a shorter time step would entail.
To illustrate the relationship between the simulation time step and the computational capacity needed, IBM researchers picked a time step on the order of 10e-15 seconds. A time step this size can accurately describe the fastest vibrations of the protein and solvent system, according to IBM.
The folding time for a protein is about 10e-4 seconds. So the simulation program must perform the small time steps 10e+11 times on every atom to simulate that set duration. That's 100 billion steps.
In other words, the computational load is huge. In an IBM research paper Blue Gene: A vision for protein science using a petaflop supercomputer, (IBM Systems Journal, Vol. 40, No. 2, 2001), the authors noted that to do a straight molecular dynamics simulation of a protein folding over 10e-4 seconds would require about three years of computational time with a petaflop computer.
Thus, one of the challenges of the Blue Gene effort is to not just to produce faster hardware, but to also develop more efficient protein folding algorithms that decrease the computational tasks while still accurately simulating the real science that is happening. To that end, IBM has been conducting global workshops that bring scientists together to share the best practices for protein folding simulation.
Back to Think Blue... Again