search

kuixu / PrismNet

Predicting dynamic cellular protein-RNA interactions using deep learning and in vivo RNA structure
MIT License
57 stars 12 forks source link

How construct data with predict RNA secondary structure and how predict specific binding location? #4

Closed eternal-bug closed 5 months ago

eternal-bug commented 1 year ago

Hello kuixu!

Thank you very much for developing this tool. After looking at the paper on the tool, there are two questions for me.

  1. How to construct the input data with RNA secondary structure(predict with RNAfold) rather than Icshape experimental data? (I read the code in GitHub and found that the training data used seq, icshape and Label. I had questions about this icshape. After reading the Icshape-Pipe paper, I couldn't understand how this number was calculated. Is it equal to the probability of secondary structure formation for each base?)
  2. How do you predict specific RBP binding location for a given long RNA?(Is the prediction generated by sliding window segments?)
kuixu commented 1 year ago

You are exactly right! 1) the icshape score represents the probability of the nucleic acid being single-stranded in the secondary structure.. 2) For long RNA, PrismNet predicts binding sites by sliding window. By the way, we provide a web server for RBP binding sites prediction using PrismNet, which could automatically deal with long RNA. Link: http://prismnetweb.zhanglab.net/.

eternal-bug commented 1 year ago

Thank you very much. I have another question. What is the width and step length of the sliding window?

Majimag111 commented 1 year ago

It looks like the online service is not available.

eternal-bug commented 5 months ago

According to the article code, sliding window width is fixed, value is 101. step is flexible.