New 3D structure prediction servers

General Science Discussion
,
1

Rhiju gave me the URL of two of the best new 3D prediction algorithms released recently. These perform better than RNAComposer. (Edit: Not exactly correct. See Rhiju’s note below.)

The first is RhoFold, which uses machine learning despite the dearth of verified 3D structures: https://proj.cse.cuhk.edu.hk/aihlab/rhofold/
The second was the best computed prediction at CASP15 for RNA: RNA structure prediction by trRosettaRNA

I tried out RhoFold. It requires the sequence in FASTA format. I’ve never created a FASTA file so I used the one provided on the website. And I used the pre-generated MSA file as a quick test to see how this works.

Screen Shot 2023-02-13 at 3.21.58 PM
Screen Shot 2023-02-13 at 3.21.58 PM978×566 64.9 KB

After clicking Submit, I was taken to a screen with this text:

Screen Shot 2023-02-13 at 3.16.28 PM
Screen Shot 2023-02-13 at 3.16.28 PM1009×594 80.2 KB

The area below was blank. After two minutes, the model appeared

Screen Shot 2023-02-13 at 3.18.09 PM
Screen Shot 2023-02-13 at 3.18.09 PM896×530 138 KB

along with the link to download the PDB file.

Apparently generating the MSA file takes a long time. I’ll try that some other day.

Screen Shot 2023-02-13 at 3.27.20 PM
Screen Shot 2023-02-13 at 3.27.20 PM973×535 66 KB

2

Really quick clarification – RNA composer, when wielded by humans, actually does better than these servers in the recent CASP15 trials.

They are the “RNAPOLIS” models, led by Marta Szachniuk’s group, in the rankings here:

https://predictioncenter.org/casp15/zscores_RNA.cgi

For more info on current top approaches to RNA 3D modeling, you can find presentations here:

including mine on the CASP RNA assessment!

3

The two issues I had when playing around in 3-D servers were the two you mentioned , time of calculation and file format. RNAComposer fits nicely into Eterna given our resources and ported into Chimera gives a 3-D view you can play around with.
What I feel could be useful in Eterna is a API from Eterna that spits out a 2-dimensional map similar to the the mini-PTC map (below) showing all the single NT interactions, distances, etc.
The computer industry is currently pivoting towards the 5th digital dimension , the spatial dimension, so slicing the 5 dimensions into 2 and 3 for easier data analysis and interpretation I would think would interest the academic world.
mini-PTC