So, can a simple machine that compares the evolution of each nucleotide position in the four competing start strings solve this problem efficiently? Can you program this?
What if we take the current string position of a nucleotide, the desired 2D secondary structure, and the next nucleotide of (four possible) to be added in the 1D string, and the current string sequence as parameters for an automated model software development initial point of reference?
We get a binary choice (in a nearest neighbor only model).
Base pair strength versus linear (loop) string trio strength.
If we start with a single nucleotide string then there are only four potential start points (eight if you use both ends of final folded structure). The evolution of all beginning string types simultaneously should ultimately result in any single best solution, if only one exists. Otherwise several solutions may present, which we know is often the actual case.
Pardon my thinking out loud.
But it seems easy enough to attempt, though I’m probably missing something.
First off, I am no pro but from what I have read on the web some do just this,
running from one end to the other 1 NT at a time. What they get are close
to a Christmas tree. And you are still left with the problem of misfolds between
one area and another one way off on the other side of the puzzle.
I like the overlay approach, JL’s program is flexible enough to allow you to choose
which overlay to choose first. My program does the same thing (a lot poorer code),
then seeing if I can chase the misfolds out of the puzzle with pattern recognition
and next removing misfolds one by one. I am months away from getting a good score but am learning much.
Thanks for your response. And I apologize for any lack of clarity.
(And I think this should probably be done as pairwise progressions down a double string.)
Granted, that so far starting at one and and proceeding linearly down the string has not yet been particularly successful. But I believe all such attempts have used only one string and usually all adenine as the starting point. And not subjected them to a simultaneous evolution and competition among all four strands which might improve the possible outcomes.
More important is that it should be something like the Piaget automated learning machine mentioned in my secondary post. So it does not stay random but learns and infers general and specific applicable rules.
The point being to learn if a linear approach of any type can be used to solve an unknown secondary shape. Or will nonlinear approaches be the only kind that can solve our puzzles? Or can we use a linear approach for some segmental size and then repeat with the segments again, in a linear fashion?
I ask the question this way because if there is a known solution you can obviously construct a machine that will linearly assemble the correct solution, even multiple solutions. So, why can it not be done with an unknown solution? It may be provable that linear approaches to the unknown case can’t work, I just don’t know that yet.
My current thought is that natural RNA forms in a combination of 5’-3’ additions and folding with occasional refolds. That is, the natural shape of the RNA as the sequence is extended is often, but not *always*, a prefix of the final target shape. A corollary is that most hairpin ends probably form before the loops or bulges or multi-loops that they are are attached to.
But natural RNA is also probably not “invented” whole-cloth, but evolved from similar shapes and sequences where a mutation brings greater stability to a shape, or a better shape (closer to the target).
So a constructive + evolutionary strategy might be as follows:
the constructive part is to start with a prefix shape that includes the first hairpin, find a solution to that, substituting ‘.’ for unmatched ‘(’, then add the next closed shape, and so on.
the evolutionary step is when a new shape cannot be added, try to mutate the prior form to avoid any conflicts, then try extending the shape again.
If it is *not* possible to form a fully valid prefix of the shape, don’t panic, but try not to get *too* far ahead before rectifying the difference. This is especially true for difficult hairpin ends that my require part of the attached loop or stack to form first, and
be willing to backtrack, changing previously set stacks when no progress can be made.
This is not how the current Random Mutations code-line works.
"This is not how the current Random Mutations code-line works. "
That was part of the reason for my thinking out-loud. My musing itself may not be that useful, but if it stimulates alternative conjectures on how the problem might be solved, then that is progress of a sort.