ScienceDaily reports that a new protein folding method may reduce the number of computations needed to predict how a protein will fold "from trillions of steps to hundreds."
If this works out, it could be huge news for molecular manufacturing.
Proteins are long stringy molecules that fold up into complex and useful shapes due to very subtle interactions between their component parts. Proteins do most of the molecular manipulation work in our bodies, joining and splitting molecules, moving things as small as atoms and as large as cellular organelles from one place to another, and making cellular metabolism work.
If it were easy to design new proteins, it would be possible to create engineered molecular machines. Put enough of the right machines together, and you could build a general-purpose (programmable) molecular manufacturing system. Even if the system could not directly build the strongest covalent solids such as diamond, it could still be used to build large-scale molecular structures, including higher-performance non-protein machines.
Predicting protein folding is a large part of designing protein machines. If you know what shape the molecule will take, you can tell a lot about what it will do. So speeding up protein folding calculations by ten orders of magnitude is a very big deal. Basically, it means that anyone with a PC (and a lot of skill) can do protein design rapidly, where it used to be time-consuming and require expensive supercomputers.
Robert Bradbury wrote a paper in 2001 about the cost and difficulty of designing an all-protein molecular fabrication system. (The web site may be down; here is an archive.org link without the pictures.) Bradbury concluded that designing the thousands of enzymes needed to assemble hundred-atom building blocks (themselves built with difficult organic synthesis) would cost about $18 billion dollars in 2001. He also estimated that the cost of enzyme design would drop by a factor of 100 by 2005, due to a variety of factors including Moore's Law, and would drop by a factor of 10,000 by 2010.
Protein folding is such an important part of enzyme design that Bradbury included improvements in protein folding as a straight divisor of the cost. That calculation may not be valid if the cost of calculating protein folding drops by a factor of 10,000,000,000. But then again, if protein folding prediction becomes trivial, automated design may become feasible, and a vast increase in data points will allow researchers to develop intuitions a lot more easily and accurately. It may be that junior researchers or even students could become protein designers after a few dozen hours of playing with a program that gave near-instant feedback of protein shape.
If Bradbury (and Freitas, whom he cites) are right that a usable protein-based molecular manufacturing system can be built with less than ten million atoms, then it would take less than a million enzymes to build the sytem out of easily-synthesized ten-atom molecules. Designing a million enzymes would be ridiculous with today's technology; in 2001, Bradbury estimated that each enzyme would take two person-years to design, and predicted that it would require about a week in 2005. But what if, with practice, it only took ten minutes to design an enzyme using the new technology? Then it would require only about 1000 person-months, or about 100 people working for a year, to design the enzymes to put together the entire molecular manufacturing system out of small molecular building blocks. (The microfluidics to handle a million enzymes, and the cost of synthesizing a million enzymes, are another question; however, the design now appears to be tractable.)
Keep in mind that the system we're talking about here is not a nanofactory and probably would not be able to make diamondoid. But it would be a major step toward those goals.