On molecular computation

Like many folks, I was fascinated with Len Adleman’s adaptation of DNA to serve as a computational entity (Adleman (1994), Science 261:1021). This was just really cool, especially in that it both encoded a problem in a nucleic acid, and then asked the nucleic acid to solve that problem based on what I like to think of as ‘hybridization logic.’ This idea quickly permeated the community, leading some to speculate that massively parallel DNA computers would be able to crack the data encryption standard that keeps our money (amongst other things) safe (Adleman et al. (1999), J Comp Biol 6:53). DNA computers (actually, RNA, to be completely wonkish) even surged to the forefront of cultural memes by being included in the Intrepid Class Starship Voyager.

So, why don’t we all have DNA computers now? There are many answers to this question, but they all sort of boil down to this: DNA computers suck. They are error prone, they don’t scale, and they are extremely poor competitors to silicon-based devices. There are of course those who would argue (correctly) that *we* are DNA-based computers, and we don’t suck. We are in fact supremely good at signal processing and pattern recognition, things that silicon computers are still catching up on. However, I would in turn argue that these features have everything to do with architecture, and not much to do with DNA. We don’t yet really know how to properly emulate a biological computer. But that said, the converse is not only true, but inherently true: I do not believe biology will ever adequately emulate silicon-based devices. Neural interconnects are not nearly fast enough, and whatever else our architectures do, they suck at calculations. Even if you take one of the savants that Oliver Sacks has popularized, their pure number-crunching capacities are much less (and much slower) than your average hand calculator. So, outside of Dune we shall not have mentats, how sad. There goes my idea for a eugenics program for savants. Probably just as well.


I always like to think of DNA computation as a good example of a failed technology arc. Initial publication, extraordinary excitement, multiple imitators, but little followup. Most of the publications in high impact journals, without much fleshing out in the second tier. A field driven by journal editors rather than by scientific utility.

And there DNA computation would have remained, except for the fact that it was intellectually … interesting. And scientists have a way of taking even the most arcane ideas, and playing with them until they pan out, even if it’s not in the way that was originally expected. From my vantage, what happened was that folks began to realize that DNA was never going to displace or even challenge silicon at what silicon was good at, and therefore it might be useful to consider how DNA computation might be useful to living systems themselves. The notion of “DNA on DNA” computation arose. And as that notion was fleshed out, information encoding became a lesser issue, and the actual silicomimetic features of the molecules themselves (to use a phrase from Milan Stojanovic of Columbia) came to the fore.

My friend Erik Winfree at Caltech would likely disagree with this interpretation, and he should be listened to: he was there for the inauguration of the field, and is at the forefront of what I like to think of as its Renaissance. He sees an unbroken intellectual lineage where I see a discontinuity. It is Erik who has more recently invented a form of DNA computation based on strand displacement (Seelig et al. (2006), Science 314:1552; Zhang et al. (2007), Science 318:1121). This model has proven to be not only intellectually compelling, but versatile. Layers upon layers of DNA circuitry can be successfully stacked. Real-world applications can be envisioned. And many lesser publications have quickly followed on the heels of the seminal ones. Moreover, the recognition that nucleic acid structural reorganization can be used as a ‘matter computer’ (in the words of Zack Simpson) has sparked new instantiations of Erik’s breakthrough, including a quite novel DNA self-assembly scheme pioneered by Niles Pierce and Peng Yin, also of Caltech (Yin et al. (2008), Nature 451:318; although Peng has now moved to Harvard).

Because of these advances, which I certainly did not anticipate but now celebrate, it is in fact possible to imagine operating systems that are symbolically encoded yet laid out in DNA. Which means that a rational biology based on programming may indeed be within our reach. It will require massive recoding, since evolution engineers by chance rather than design, but it is nonetheless for the first time more than just a science fiction story.


- originally posted on Saturday, November 6th, 2010