Some issues about 2d prediction

[penglbio]

Hello Anton Petrov,

I would like to predict the 2D structures of a large number of archaeal RNase P RNAs. I have downloaded and installed the Docker package locally. Using the RNA sequences of archaeal RNase P RNAs from your website, I ran the following command:

r2dt.py rnasep draw temp/mydata/archaea_RNaseP_rna_4190.fasta temp/archaea_RNaseP_rna_4190

I provided over 4000 sequences, but only 874 of them successfully generated 2D structure. However, if I force to specify a template explicitly, the prediction works and structures can be generated.I ran the following command:

r2dt.py draw --force_template RNAseP_a_P_furiosus_JB temp/mydata/archaea_RNaseP_rna_4190.fasta temp/force_4190_RNAseP

Even though there are some errors happened，but more than 4000 of rna generated 2D structure.

I also tested this issue on the r2dt tool of web version, and the same problem occurs there. 

As a specific example, I tested the sequence with the ID URS0002213FCB_2026803. It fails to produce a 2D structure unless a template is manually specified.

Actually, I don't understand why specifying a template enables successful prediction, but using all_templates fails. When predicting structures in batch mode, I would prefer the tool to compare all templates and select the best-matching one automatically. I also noticed that when I manually specify a certain template, the predicted structure sometimes differs from the one previously shown on the website. Moreover, I feel that the prediction results I get are not always of good quality. and I even suspect that the error causing the image not to be displayed on the website might be related to this issue.

Could you please help me understand what causes this issue and how to ensure more sequences can be processed successfully?

Best regards,Lily

Originally posted by @penglbio in #25

[AntonPetrov]

👋 @penglbio Hi Li,

Thank you for following up! I’ve pushed a fix as a separate Docker image tagged pr-194. Please give it a try — you can run it without specifying a template, and hopefully it will behave as you expect.

docker pull rnacentral/r2dt:pr-194
docker run -v `pwd`/r2dt/temp:/rna/r2dt/temp -it rnacentral/r2dt:pr-194
r2dt.py draw temp/input.fasta temp/output

The main issue stems from a change in version 2.0, where we introduced a BLAST pre-filter to speed up template selection. For some of your RNAse P sequences, this BLAST step fails to find matches in the template library. However, a slower but more sensitive method does identify the correct templates. In pr-194, I’ve disabled the BLAST pre-filter specifically for RNAse P, so all your sequences will now go through ribovore, which should select the appropriate template.

I suspect that forcing a template may have led to mixed results because there are many RNAse P templates, and the one I previously suggested might not have been optimal across all 4,000 sequences. By allowing the system to automatically select the best template, my hope is that some of the crashes you’re seeing will be resolved.

Please give it a try and let me know how it goes. Thank you again for reporting this — it’s been very helpful!

[penglbio]

I’ve tried it. The result is perfect for me, but I’d still like to know how this small error in the output was generated. thanks a lot.

[AntonPetrov]

Glad the fix worked! I will look into the enrich_json problem and get back to you asap.

[AntonPetrov]

@penglbio Thank you for reporting this issue! I’ve investigated further and found that the problem lies with one of the programs used by R2DT.

@davidhoksza Hi David! I tracked this issue down to a discrepancy in the JSON file generated by Traveler and the TSV file with nucleotide-specific annotations.

In sequence URS00028A3AF5_3110270, Traveler produces a JSON file where C 176 is missing, even though it is present in both the input FASTA and the TSV file. As a result, the enrich_json.py script crashes.

Below, the first sequence line comes from the JSON file, while the second and third lines are from the FASTA and TSV, respectively:

Here are the relevant files:

URS00028A3AF5_3110270-RNAseP_a_H_volcanii_JB.colored.json
URS00028A3AF5_3110270_post_prob.txt

URS00028A3AF5_3110270.fasta

>URS00028A3AF5_3110270
GGCAGAGAGAGCCCGGCCUCCGUGAACGCGCCGCUGGCGCGUGCAUGAGGAAAGUCCCCCACCGUCCGGGCAGGUGACCGGGCGCAAGCCCGGGGCGGGAGACCGUCGGCUAUGGAACAGAAACGAGACCACUCCACCCGACCGAUGACGUGCGCGAACCCGAACCGAGAAGGCACGGGCCGUCCAGGCGCGAGCCGAGAAAGGAACGGGGAGUCACCCACAGAGGGCCGACGGUGGAGAACGGAUGGAACGGCGAAUCCUCACCGGUGCAAGCCCGCAGGGUAGGUAGCCCGGAGGCACGUGGAAUUCCACGCCCCUGCGGACGCUCAGCCGAAUGCCGGGCCGAACAGAAGGGGCUUACUCCUCUCAGCC

Commands to reproduce the crash (note: this is based on the R2DT code from PR #194, which fixes a template selection bug but does not affect Traveler):

docker run -v `pwd`:/rna/r2dt  -it rnacentral/r2dt:pr-194
r2dt.py draw temp/URS00028A3AF5_3110270.fasta temp/URS00028A3AF5_3110270

Would you be able to look into this, when you have time? Many thanks in advance!

[penglbio]

thanks for your help again.

[penglbio]

I have another question. I currently have a DBN file, and I used template-free mode to generate a secondary structure JSON file. However, the resulting structure diagram isn't very good. So, I'm thinking that using this JSON as a template to generate other structures may not work well either.

Therefore, I wonder if I can first edit the secondary structure JSON file, for example, by adjusting the positions of overlapping bases, resizing loops, etc., and then use the edited JSON to generate a new template. Would this improve the prediction results?

I'm not sure if this is feasible. Also, is there any software that allows editing the JSON and still outputs a valid JSON file afterwards?

[AntonPetrov]

Just a quick response on the go to say that yes, you can edit the diagrams and then use them as templates:

https://docs.r2dt.bio/en/latest/editors.html
https://docs.r2dt.bio/en/latest/templates.html

I will address the other points in more detail shortly.

Hope it helps!