Dear Modellers,
I found some interesting results today.Though some of these were much discussed by Sadowsky in some articles since 2007,
But still check these out.
Target1
molpdf DOPE Z-Score GDT-TS GDT-HA TM-Score RMSD
2P31 Modeling Scores BLOSUM50 924.97821 -19218.93945 -1.37493 89.634 76.22 0.917 0.844
BLOSUM62 924.97821 -19218.93945 -1.37493 89.634 76.22 0.917 0.844
2I3Y Modeling Scores BLOSUM50 1027.70447 -18048.86914 -0.89869 79.268 57.165 0.873 1.568
BLOSUM62 1027.70447 -18048.86914 -0.89869 79.268 57.165 0.873 1.568
2OBI Modeling Scores BLOSUM50 1019.33649 -18713.56836 -1.16924 85.671 68.293 0.914 1.325
BLOSUM62 993.73724 -14176.32422 0.67751 39.482 15.244 0.586 .3.727
Target2
2RIR molpdf DOPE Z-Score GDT-TS GDT-HA TM-Score RMSD
BLOSUM50 1528.87537 -34527.82422 -1.28434Complete89.568 68.255 0.973 1.037
Domain1 96.507 83.824 0.972 0.77
Domain2 99.823 89.184 0.985 0.553
BLOSUM62 1528.87537 -34527.82422 -1.28434Complete89.568 68.255 0.973 1.037
Domain1 96.507 83.824 0.972 0.77
Domain2 99.823 89.184 0.985 0.553
2CUK molpdf DOPE Z-Score GDT-TS GDT-HA TM-Score RMSD
BLOSUM50 1942.12292 -25812.05664 0.82958 Complete17.356 5.665 0.382 4.106
Domain1 25.735 13.419 0.361 2.613
Domain2 30.674 13.121 0.411 2.93
BLOSUM62 2036.98694 -26931.20703 0.55814 Complete16.277 5.036 0.348 4.12
Domain1 15.625 6.066 0.256 2.82
Domain2 34.752 15.426 0.456 2.747
Target3
molpdf DOPE Z-Score GDT-TS GDT-HA TM-Score RMSD
2Q7B
BLOSUM50 1212.10278 -16748.72461 -0.8108 81.414 61.184 0.879 1.283
BLOSUM62 1175.47266 -17014.03516 -0.92678 87.336 66.118 0.925 1.227
2OH1
BLOSUM50 1450.15808 -12735.95215 0.94347 28.947 13.487 0.426 2.797
BLOSUM62 1007.18695 -15380.72852 -0.21275 56.743 33.388 0.73 2.367
1S60
BLOSUM50 836.13391 -13656.52148 0.54102 55.428 35.526 0.674 1.961
BLOSUM62 1039.06799 -13509.62109 0.60524 57.237 37.500 0.698 2.002
So You might have realized there is a huge differences in the alignments. Sometime same and different for quite others.
Even the accuracies, molpdf and other scores are different, when it comes to the models created on basis of only this factor keeping all others constant.
Even the accuracy of MODELLER single template models also gets inferior sometimes.
So I infer that it seems
1. MODELLER pairwise algorithm is not purely biased on local fold alignment, though it uses Smith Waterman based align2d.py module.
2. All algorithmic alignments are almost similar with minimal differences for easy targets. However, when the PDB profile is employed for Pairwise alignments through any BLOSUM matrix, the case may differ much when the employed template shares a negligible sequence identity to the target.
3. Will these results be still different when BLOSUM80 is used for these targets the same way?
I hope a good explanatory reply from Sir Ben Webb.
Thanks
Ashish
Ashish Runthala,
Lecturer, Structural Biology Cell,
Biological Sciences Group,
BITS, Pilani
Rajasthan, INDIA