I would think that a well behaved bot should be able to be constructed just using these tables and with no math calculations (they’re embeded in the table construct) it might be quite fast. Further these tables would allow all known non-Watson-Crick pairs and all expected pairs generally.

It would be way more comprehensive than our current model, if I’m not off base.

Eternacac, this looks like an interesting article. Thanks for pointing it out.

As for you original question – I don’t know the literature, but speaking as an algorithm designer, it doesn’t seem likely. Here’s a brief explanation as to why I say that.

Finding a “good” sequence requires both local knowledge (e.g. how well does a specific base fit with its adjoining/paired bases) and global knowledge (e.g. is there a plausible way for the RNA backbone to thread all the bases together). So any solver is likely to rely on some variation of a “generate and test” strategy. That is, it will repeatedly generate a (perhaps partial) solution and then test it to see how promising it is. In the big picture, the “test” part is generally cheap, in the sense that doubling the calculation associated with each test “only” doubles the whole run time. But even a small increase in the number of possibilities considered at each stage of the generation process increases the overall solution time exponentially. For example, if there are 100 bases in the target structure, and each step in the generation process consisted of adding one more base, doubling the number of bases generated at each step would increase the overall run time by a factor of 2 to the 100th power (which is roughly a 300 digit number.)

According to the article, there are six degrees of geometric freedom between any two pairs of bases. (That is, it takes 6 floating point numbers to describe their relationship.) Given that as a minimum, (almost) every base is heavily influenced by two adjoining bases, and many are heavily influenced by a third (a paired base), there are probably something like 16 degrees of geometric freedom (on average) associated with each base’s relationship to its neighbors.

My conclusion is that while geometric information might well be useful in the “test” part of a generate and test algorithm, it’s not a very promising candidate for the “generate” part.