@LFP6, no I haven’t shared known to work sequences that RNNet doesn’t predict well. I just noticed that some of the lab puzzles - also some with crystal origin - weren’t stable in RNNet with their original sequence.
But to answer your question, here comes some examples I have just added to Omei’s and mine spreadsheet Non-canonical bonds in pseudoknots
Hey all - a bunch of updates since we last talked! The latest version is no longer at the demo link above, but instead it is now available with further updates at https://eternadev.org, and I plan on releasing to the main site tomorrow.
Some included updates:
Some performance improvements (centered around converting to and from dot-bracket strings with pseudoknots), as well as a significantly lower download size (this is an optimization separate from RNNet)
We now support pseudoknots of order higher than 3, and we explicitly error if we encounter situations we don’t support (eg, PK orders too high for us to represent). We use a-z matched with their uppercase partners (A-Z) to get another 26 orders.
Adjustments to eF1 according to the latest RibonanzaNet updates
RNNet is lazy-loaded (we don’t download it until you actually chose to use it)
The dot plot now no longer draws values <0.001, which caused the entire thing to appear gray (note that we draw everything under 0.1 as 0.1)
Consistently disable the free energy displays when unavailable
Support async versions of a number of methods in EternaScript, and add safety checks to ensure the game isn’t modified when it’s at a point where it’s only partially updated
Use square brackets ([]) before curly braces ({}) when generating dot-bracket strings (this aligns with Arinie and most other packages - previously we used NUPACK’s convention)
Various (largely internal) cleanups and bugfixes around pseudoknot handling, locking the UI while processing, and keeping various parts of code in sync and not running out of order
Two additional misc notes:
I think there are potentially ways to get more accurate eF1 scores - I’ll be talking to Shujun about this at some point, so be aware that this metric may change
Unfortunately, it doesn’t look like we can run RibonanzaNet multithreaded, which means it will be slower. (Due to some browser security limitations, we can’t have YouTube/Vimeo embeds AND turn on multithreading. This is due to how YouTube and Vimeo are configured, and I would argue their configuration is incorrect )
)