So here I am at this conference on self-assembly, and I'm talking with people about Rothemund's DNA staple technique. The technique is elegant, simple, and powerful. Mix a long DNA strand with a bunch of short strands. The short strands bond to different regions of the long strand, and pucker it up into a shape.
Person after person has told me that they never would have expected this to work! I was really surprised at that, so I've started having conversations about it.
One of the most common reasons has to do with entropy. Perhaps it's not a coincidence that entropy is one of the most common objections to molecular manufacturing. There's a whole lot of very solid theory about entropy, which says that unstable molecules will fall apart. But what's often missed--by the people applying this theory!--is that it assumes a steady state, also known as infinite time! In other words, entropy says diamonds aren't literally forever... but in the real world, we can ignore that. Nanomachines built of diamond could laugh at entropy for far longer than we actually care about.
So, back to entropy vs. DNA self-assembly. I was talking with a physicist over lunch today, trying to figure out why he expected DNA shapes to fall apart on entropic grounds. It turned out that he was right - the shapes will eventually fall apart - but he hadn't thought about the time it would take.
Another entropy-related reason that's often cited is that DNA staple shapes require combining hundreds of molecules into a single structure, and the entropy of a single structure is a lot lower than the entropy of hundreds of fragments. While this is true, the fact remains that DNA does prefer to join in double strands. I haven't found anyone who can explain why increasing the number of strands would significantly reduce the willingness of each strand to join the structure - but I've found lots of people who seem to believe this would be the case.
I can understand how assumptions from one domain (say, the physics of small molecule chemistry at equilibrium) can make it difficult to invent a new concept in another domain (say, DNA binding). It's a bit harder for me to understand how that assumption wouldn't be questioned when confronted with a really good idea.
In fairness, it's true that many seemingly good ideas don't work for reasons that couldn't easily be foreseen. So rejecting a seemingly good idea isn't always a destructive thing to do - it can save a lot of effort that would be wasted trying to develop the idea only to discover that some minor practicality prevented it from actually working.
But I'm starting to think that the stated reason for rejecting the idea is frequently unrelated to the reason that the idea wouldn't actually work. In other words, science has developed a rule of thumb that works: Reject most ideas, accepting only the most compelling and fully demonstrated ones. And a procedure that works: To reject an idea, find and cite some overly-generalized and poorly understood theory. The fact that the procedure is groundless does not mean that science's rate of rejection of new ideas is wrong. The rate of rejecting new ideas may, in fact, be highly functional.
If the rejection of new ideas is basically random, then I'm saying that science advances by blind luck. In fact, that's plausible. Any exploration of a sufficiently unfamiliar and difficult problem domain must advance by blind luck. What makes science (and evolution) work is that they can, after the fact, detect and preserve lucky accidents.
A question we might ask is whether things would be better with a somewhat lower idea rejection rate, and if so, how to shift the rate. The answer is probably somewhere along the lines of increasing funding for researching crazy ideas. I don't know if there would be any point in trying to convince scientists that they're routinely rejecting ideas randomly.
Another question we might ask is what to do when scientific advances are needed to achieve a technological goal. The answer seems to be: Don't expect the scientific establishment, as a whole, to work toward that goal, no matter how workable or clearly articulated or well-calculated. The goal will simply be treated as a new idea, and rejected. Instead, find a way to pay individual scientists to work on it.
By the way, I just heard a talk in which a DNA cascade circuit was developed and tested to compute the square root of a four-bit number. It involved about 130 strands, implementing about 76 logic gates. It would be very easy to find a theoretical reason why this couldn't possibly work... but in fact, it did work.