MakeTNT-toolkit User Documentation To MakeTNT-manual
MDL2TNT/PDB2TNT Evaluation

Comparative Test Runs for MDL2TNT and PDB2TNT

Evaluation :  Generation of Restraints by MDL2TNT

Source of compounds Number of compounds Successful Generation of Restraints Failures
NCI diversity data set  1990 1990 -

Evaluation :  Accuracy of BOND recognition by PDB2TNT

The bond recognition algorithm of PDB2TNT has been tested using 3D-coordinates (but not the bond table !!) from the NCI diversity data set and bond-restraints generated this way have been compared to the ones obtained by MDL2TNT (with explicit knowledge of bond orders). Results are shown below:

Source of compounds Number of compounds Number of structures with fully correct bond restraints Number of structures with missing bonds Number of structures with too many bonds Reasons for additional bonds
NCI diversity data set 1990 1970 - 20
  • bumps in input structure (12 compounds)
  • too large tolerance for bond distances in metal complexes (8 compounds)

Evaluation :  Accuracy of PLANE Restraints by PDB2TNT

Recognition of planar restraints by PDB2TNT has been tested by comparing planar restraints (format:s) obtained by its hybridisation-determining algorithm to the ones obtained from MDL2TNT. 2 test runs have been performed: one including the positions of H-atoms and one without knowledge of H-atom positions. Manual inspection of discrepancies between MDL-file and PDB-file derived restraints indicated that:

Source of compounds Number of compounds Inclusion of H-positions Differences between MDL- and PDB-derived restraints Reason for differences:
NCI diversity data set

3D-MDL-files

1990 yes 34
  • 24 cases: planar bond in non-planar 5-membered ring
  • rest: double bonded sulfur, metal coordination etc.
NCI diversity data set

3D-MDL-files

1990 no 83
  • 41 cases: planar bond in non-planar 5-membered ring
  • 29 cases: planar bond not recognised because of strains in bond angles
  • rest: double bonded sulfur, metal coordination etc.

Evaluation :  Small Molecule Refinements Using restraints Generated by MDL2TNT and PDB2TNT

Each cell in the table below represents one optimisation run, using the TNT-GEOM module without X-ray terms and a fast optimization routine kindly provided by Dr. Oliver Smart, Globalphasing.

Initial coordinates (with the correct stereochemistry around asymmetric centers) have been randomized to RMS deviations of 0.5 and 1.0 Ångstrøms and optimized with TNT-dictionaries, sizes m and l, derived either from a 3D-mdl file (M) or a PDB-file (P). For example, 0.5/m/M means, the compound has been distorted to an initial RMSD of 0.5Ångstrøms, optimized with a TNT-dictionary file of size m derived from a 3D-MDL file.

Optimizations producing apparent errors in the geometry have been marked in red. The results indicate that:

compound
number of atoms
structure optimisation: jiggle[Å]/Size of -Restraint file/Source M:MDL2TNT; P:PDB2TNT
no. refinement cycles/successful?
0.5/m/M 0.5/s/M 0.5/m/P 0.5/s/P 1.0/m/M 1.0/s/M 1.0/m/P 1.0/s/P
NCI124652/28 98
yes[a]
116
yes[a]
83
yes[a]
40
yes[a]
91
yes[a]
101
no
98
yes[a]
71
no
NIST04051410392D/10 36
yes
36
yes
45
yes
45
yes
50
yes
50
yes
106
yes
106
yes
NCI16498/25 44
yes
79
yes
56
yes
46
yes
80
yes
71
yes
79
yes
48
yes
NCI367798/48 120
yes
267
yes
115
yes
87
yes
115
yes
215
yes
91
yes
113
yes
NCI376722/29 96
yes[a]
98
yes[a]
75
yes[a]
58
yes[a]
101
yes[a]
156
yes[a]
102
yes[a]
91
yes[a]
NCI46/20 65
yes
107
yes
77
yes
92
yes
92
yes
148
no
103
yes
91
no
NCI56940/28 84
yes
98
yes
87
yes
63
yes
105
yes
116
yes
124
yes
104
no
NCI628503/58 104
yes
118
yes
81
yes
101
yes
109
yes
1091
no
115
yes
323
no
NCI631303/29 80
yes
154
yes
89
yes
78
yes
76
yes
91
yes
96
yes
113
yes
NCI640266/26 152
yes
78
yes
154
yes
57
yes
132
yes
73
yes
158
yes
79
yes

[a]: structure of good quality, except for positioning of metal ions(e.g. Cu, Co), where currently no parameters for non-bonded interactions are available in Buster/TNT


Maria Brandl,<buster-develop@GlobalPhasing.com>
Last modification: 13.12.05