I think that i found a problem with the bot algorithm. They do not handle big loops properly. I tested this in the player puzzles Easy Bot Test, big, huge, and huger!, and i have made another puzzle that the bots have not yet commented on. most of the big loop puzzles that i was testing them on were puzzles that were easy to solve – you just put the CG pairs near the ends of the stacks. if this is not the right place to bring this up, then pleas tell me where to post it.
It does seem to be a pattern that the bots don’t like big loops
This is really interesting. There are other features of designs that are hard for bots, like short stacks, etc. See, for example:
https://getsatisfaction.com/eternagam…
There is also the zigzag feature, and other things that make designs hard.
There has actually been no paper analyzing what makes an RNA design problem hard – for bots or for humans. We’re thinking of writing one – hopefully we can contact you to get more insights! Please continue to record observations here.
I am testing how big the loops need to be in order to be “to big” for each bot now. I am starting at the loop being 35 nucleotides long, and increasing or decreasing how big the loop is by 5 each time. So far none of the bots have replied, but i believe that 35 will be to big for most of them.
So far this is the data for the bots being unable to solve large loops. this was tested with a hairpin loop that had a stack length of 4.
Vienna - times out at 40
InfoRNA - times out at less then 35
RNASSD - I have not gotten RNASSD to time out yet.
NOTE: When i say “times out at 40”, i mean that it could not solve a hairpin with a loop 40 nucleotides long and a stack 4 long.
I did a series of puzzles titled ‘Infovienna tests’. What I discovered is that bots preform worse on RNAs where the mid point or end loop of a strand or hairpin shape RNA is clockwise of the origin. I also discovered that the bots suffer when there are many unpaired bases. I wrote in greater detail about these findings here: https://getsatisfaction.com/eternagam…
I hope you find this information helpful.
Thank you for the information. i will look into this more.
Hi Nsmurf!
I like that the pattern you found continues. Brourd’s puzzles Just another small puzzle 14 and 15, have loop sizes of 76 and 83 nucleotides. And RNASSD is the only bot not failing to solve them.
I have been collecting some more data, and ere is what i have discovered:
After reading up more on Vienna, it has a semi random solution, so it has a specific chance of failing that will remain constant on all of my bot tests so far.
InfoRNA cannot do a test that has more then 35 nucleotides.
and i have not gotten RNASSD to time out, so i will change my strategy: I will start at about 100 nucleotides and work my way down.
I immediate liked this approach. When I uploaded the word list to Rhiju’s and JP Bida’s paper, I fully realized what you were doing. You were taking a heuristic approach.
Definition from JP Bida:
HEURISTIC - Rules for solving a problem based on experience and not necessarily physics. For example, if you want to hit a castle with a catapult you could use the heuristic rules.
- Fire the catapult
- If you fired too far raise the pin
- If you fired too short lower the pin
Good call, Nsmurf, fire away!