Difference between revisions of "Talk:Sequence Alignments Gaucher Disease"

From Bioinformatikpedia
Line 6: Line 6:
 
* I'm not sure about your evalutation of the hit list in pdb: You only report the overlap but don't use them as test set?
 
* I'm not sure about your evalutation of the hit list in pdb: You only report the overlap but don't use them as test set?
 
** To which extend should we use the PDB hits as test set? Do you mean that we should have done a ROC analysis? As I wrote in the wiki, we would have needed a uniform test set for HHblits and blastpgp for being able to compare the different methods thoroughly. Another point is, that I do not really known how to classify the PDB hits as true positives and false positives. Did you find a good solution for that?
 
** To which extend should we use the PDB hits as test set? Do you mean that we should have done a ROC analysis? As I wrote in the wiki, we would have needed a uniform test set for HHblits and blastpgp for being able to compare the different methods thoroughly. Another point is, that I do not really known how to classify the PDB hits as true positives and false positives. Did you find a good solution for that?
  +
*** Not really, it just sounded like you had one for PDB but didn't tell us. :)
 
* If I get the task description right, we are supposed to create 5 datasets for the alignments, as you have done. Only if this is not possible, 3 less strictly defined datasets can be used.
 
* If I get the task description right, we are supposed to create 5 datasets for the alignments, as you have done. Only if this is not possible, 3 less strictly defined datasets can be used.
 
--[[User:Boidolj|Boidolj]] 18:38, 7 May 2012 (UTC)
 
--[[User:Boidolj|Boidolj]] 18:38, 7 May 2012 (UTC)

Revision as of 20:29, 7 May 2012

Just out of interest (I'm not reviewing your page this week): How did you generate your Venn diagrams? I really like the way they look.

  • We made use of this online tool.

I see, you're not finished with the alignment part yet, but a few comments:

  • I'm not sure about your evalutation of the hit list in pdb: You only report the overlap but don't use them as test set?
    • To which extend should we use the PDB hits as test set? Do you mean that we should have done a ROC analysis? As I wrote in the wiki, we would have needed a uniform test set for HHblits and blastpgp for being able to compare the different methods thoroughly. Another point is, that I do not really known how to classify the PDB hits as true positives and false positives. Did you find a good solution for that?
      • Not really, it just sounded like you had one for PDB but didn't tell us. :)
  • If I get the task description right, we are supposed to create 5 datasets for the alignments, as you have done. Only if this is not possible, 3 less strictly defined datasets can be used.

--Boidolj 18:38, 7 May 2012 (UTC)