I’ve been spitballing some puzzle-solving ideas lately and I was told that my idea of natural-target delta minimization of the confidence parameter (RNA SPEC window) might be interesting enough to even warrant a sort of confidence delta interface metric in a later update. I was told that I should make a post here for this reason.
I believe this could be used as an alternative to energy delta minimization, but I’ve been sluggish in my experimentation. It would be nice if I had a script to calculate the delta for me, but I have next to no experience with coding.
Hey Aaron! What would the delta be between? In the case of free energy, the goal is to make the target mode the lowest energy of competing structures, hence calculating the delta between the target energy and native/minimum free energy. Just lowering the target energy doesn’t necessarily lab to improvement. However, it is meaningful to just try and increase the target confidence - a target confidence (BPP/F1) of “1” should (IIRC) mean the dot perfectly matches the target, which means there are no other possible structures and hence no more room for improvement.
I’ve been trying to minimize the delta between natural and target confidence as shown in the RNA spec window (not the PK confidences yet). To be fair, I don’t even know much about what this metric even is. I’ve yet to see it equal 1 in my solved structures so I assumed it simply measures the confidence of the highest-confidence structure.
If the natural structure is the same as the target structure, there is of course 0 delta between natural and target confidence. Minimization then functions similarly; each change you keep will encourage the target confidence and discourage the most confident competing natural structure in relation. Unlike energy delta minimization, there are of course many more natural structures that can have the same confidence as the target and thus you will run into more 0 delta dead ends when using it as a solving tool. However, I have found that this approach is at least good at finding solutions composing of similar base pair compositions but of different orientations and positionings in stacks. Thus, this is at least good for finding variations of solutions.
The main strength, and really what I’ve been concerned with, is that this works on threshknot puzzles where energy delta minimization cannot be done. Currently I’m thinking about the PK confidence. I’m not really sure what this actually is, but it seems distinct yet behaves similarly to the other confidence where its delta equals zero on a solved structure. If this is true to some extent, the minimization of the sum of these confidence deltas could make for a solving strategy that does not run into 0 delta dead ends as often.