Denovo prediction from multiple fragment lengths

2 posts / 0 new
Denovo prediction from multiple fragment lengths
#1

I read that a combination offragment lengths is best for prediction. I tried to cat different fragment legnth and gave it to Rosetta, but it complained:

caught exception

[ERROR] EXCN_utility_exit has been thrown from: src/core/fragment/ConstantLengthFragSet.cc line: 139
ERROR: aframe->length() == max_frag_length()

caught exception

[ERROR] EXCN_utility_exit has been thrown from: src/core/fragment/ConstantLengthFragSet.cc line: 139
ERROR: aframe->length() == max_frag_length()



In a supervised approach, we should be able estimate the rough length of the secondary structure elements from the topology and choose fragment lengths that would correspond to it, no?

Post Situation:
Sun, 2019-03-24 03:55

So, as you might have surmised from the error message, the problem you're running into is that you're reading things into a ConstantLengthFragSet, which assumes that all the fragments are of the same length. This is the typical way of handling fragments in Rosetta - you have two sets of fragments: long fragments (normally 9-mers) and short fragments (3-mers).

If you go against that assumption, you're likely in for a bit of mucking around to get things to work, as some things are going to be assuming you're working with a standard scheme. -- That doesn't mean you can't work with other fragment schemes, but it might take you a bit of effort to get things to work.

I do believe that people in the Baker lab have played around with some different fragment schemes, including adding longer fragments or having variable length fragments. That said, I don't know how much of that has made it into the distributed version of Rosetta. For the most part, these alternate fragment schemes (at least the ones tried by the Baker lab) have not shown a substantial improvement in results.

If you do want to play around with it, you're likely going to need to do a bit of code diving to see how the cogs and gears work. My recommendation is to first take a look at the fragment file reading code in main/source/src/core/fragment/FragmentIO.cc  - This already has support for three different types of fragment sets, depending on the format of the file. You might find that one of these formats is better for your variable length approach. Or, getting variable length fragment support might require making a new type of FragmentSet.

Sat, 2019-03-30 12:10
rmoretti