Questions about rosetta fragment library ranking and fragment conformation sampling algorithm

[sokrypton]

Hi, I have some questions about fragment library used for de novo folding and loop modeling.

From what I have read, the score for ranking fragment conformations in a library consists of two parts, which are from psi-blast results and secondary structure prediction results. My questions are:

1) When performing the psi-blast that contributes to the ranking of fragment conformations in a library at one position, does the psi-blast involve the whole protein sequence or just a local region around that position?

2) If the query sequence doesn’t have any homologs so that psi-blast doesn’t work, how does the program rank fragment conformations? Is it then totally dependent on the secondary structure prediction?

3) The default size of fragment library at one position is 200. Does it mean in a de novo folding, all 200 conformations will be randomly selected with equal probability? And why in the Carol RoHl et al. METHODS IN ENZYMOLOGY (2004) paper, it says only the top 25 fragment conformations will be randomly selected?

Thanks a lot!!

[Anonymous]

1)PsiBlast is used to derive sequence profile for the query. The whole query sequence is involved in PsiBlast search. The query profile is prepared in the same manner as a sequence profile for each sequence in the database (the source of fragments) had been made.

When it comes to rank fragments, each residue position is treated separately, i.e. a fragment score is a sum of per-position components. E.g. for nine-mers it is a sum of 9 values describing the fit between a profile column (20 probabilities) from the query and a column from a database entry. The positions within a fragment are treated independently.

2) If there are no homologous sequences, then the sequence profile is just the query sequence. Performance of PsiPred is obviously worse. Fragment picker uses three independent secondary structure predictions, which may help getting the right secondary structure prediction

3) I am not sure, but as far as I remember Rosetta uses all 9-mers in stage 1, then only top 25 ninemers. Always ALL 3-mers are used.

Best,
Dominik

[sokrypton]

Thank you Dominik!
That is very helpful!

In you answer to 3), I’m wondering why Rosetta only uses top 25 ninemers after stage 1. Is it purely for efficiency? Will it reduce the diversity of the fragment conformations being sampled?

Also, is there any further documentation for Fragment picker in the newest Rosetta version? I remember in some previous posts, someone mentioned new picker can deal with various size of fragments besides 3mer and 9mer.

Thanks!

Yizhou

[Anonymous]

Hi,
apparently Rosetta always uses the top 25 9-mers. I presume this is the result of ab-initio benchmark. You can however change the default behavior using the -abinitio::number_9mer_frags option, say -abinitio::number_9mer_frags 100 and see what happens in your particular case.

> Also, is there any further documentation for Fragment picker in the newest Rosetta version?
> I remember in some previous posts, someone mentioned new picker can deal with various size of fragments besides 3mer and 9mer
Yes, the new picker allows pick fragments of any arbitrary length; moreover the score function that rules the selection process is fully customizable. I am working to include it in the very next Rosetta release.

Dominik

[Author]

Posts