Michael H.W. Weber
07-17-2002, 06:32 AM
Please correct if I have missed something. :D
From the "whatsnew.txt" file it is evident that, with the new CASP5 target, the scoring function - which is used to select the presumably best structure model from the bulk of generated protein structures - has been optimized:
"- pseudo-energy now uses EEF1 solvation term instead of 'crease' energy; this takes a little longer to compute but we have found is more accurate and reliable, hopefully leading to better structure predictions for CASP"
After the last update, the following three new files appeared in my DF work folder:
param19_eef1.inp
solvpar.inp
toph19_eef1.inp
I assume that these are the basis for the scoring/selection of the generated structures of which ONLY the best will be uploaded. The fact that this function had to be optimized implies that the preceding function was suboptimal. :rolleyes: Since the selection of the best structure to be uploaded to the DF server is carried out on the local client machine, I must conclude that all past results generated by DF throughout this CASP5 experiment are consequently suboptimal.
The good news is, however, that obviously we have learned something important about scoring functions. Also, I am still hopeful that the possibly suboptimal structures generated for the previous CASP5 targets are still better than those that will be submitted by the competing work groups. :D
Therefore my question: On the average (if one can say so), how much difference in each of the structure models can be expected when comparing those selected by the old scoring function to those that will be selected by the new one?
All the best,
Michael.
From the "whatsnew.txt" file it is evident that, with the new CASP5 target, the scoring function - which is used to select the presumably best structure model from the bulk of generated protein structures - has been optimized:
"- pseudo-energy now uses EEF1 solvation term instead of 'crease' energy; this takes a little longer to compute but we have found is more accurate and reliable, hopefully leading to better structure predictions for CASP"
After the last update, the following three new files appeared in my DF work folder:
param19_eef1.inp
solvpar.inp
toph19_eef1.inp
I assume that these are the basis for the scoring/selection of the generated structures of which ONLY the best will be uploaded. The fact that this function had to be optimized implies that the preceding function was suboptimal. :rolleyes: Since the selection of the best structure to be uploaded to the DF server is carried out on the local client machine, I must conclude that all past results generated by DF throughout this CASP5 experiment are consequently suboptimal.
The good news is, however, that obviously we have learned something important about scoring functions. Also, I am still hopeful that the possibly suboptimal structures generated for the previous CASP5 targets are still better than those that will be submitted by the competing work groups. :D
Therefore my question: On the average (if one can say so), how much difference in each of the structure models can be expected when comparing those selected by the old scoring function to those that will be selected by the new one?
All the best,
Michael.