Hi Eterna,
I organically decided to write this post as I could use some help soon and wanted to make sure everyone understands the limitations of pknots in nupack code.
So to start, if you read through the papers that the NUPACK team wrote to back up their model they talk about how pknots come in varying configurations and they expressed concern that it was very difficult to develop a one size fits all algorithm for pknots so they developed a model that only allows a very narrow subset of pknots based on commonality in nature. As such the nupack pknot routine is very limited in that it is able to process every pknot but it does not know how to know it has a pknot to process efficiently.
I have been working with LFP6 to go over the issue and I digested a large portion of the papers and code to verify their code matches their paper as we thought the limits were bugs… Eventually I was able to confirm that everything looks right and it matches what they say is happening . The problem is that we need this routine to do pknot research and it does not work on some of the stuff in the pipeline.
The solution right now is that I have been working on a new algorithm to replaces the NUPACK teams and i think I have it done. Problem is I dont think that the Nupack team gave up really quick so maybe there is a gotcha I missed and anyways we need to really test this thing out hard when I get it working…
That being said I would love it if I could get some volunteers to join this project who can help with coding or testing coding and is able to give feedback. I have a github project for this and it would all be done through that . Respond to this if you want to help or you can also say something in the slack folding-engine channel.
Hope everyone has a good weekend!
Github link for project