MaxSub score of 9: T0047AL066_1 |
MaxSub score of 6: T0082TS136_1 |
 |
 |
Model no. 1 from group 066 on CASP3 target T0047.
158 residues in the target.
Superposition of all 157 residues of the model: 1.25 A
MaxSub well predicted subset size: 152,
superposes with an RMS of 0.91 A.
|
Model no. 1 from group 136 on CASP3 target T0082.
190 residues in the target.
Superposition of all 190 residues of the model: 5.93 A
MaxSub well predicted subset size: 133,
superposes with an RMS of 1.75 A.
|
MaxSub score of 4: T0081AL212_1 |
MaxSub score of 4: T0061TS185_1 |
 |
 |
Model no. 1 from group 212 on CASP3 target T0081.
151 residues in the target.
Superposition of all 100 residues of the model: 3.04 A
MaxSub well predicted subset size: 81,
superposes with an RMS of 1.76 A.
|
Model no. 1 from ab initio group 185 on CASP3 target T0061.
76 residues in the target.
Superposition of all 76 residues of the model: 8.94 A
MaxSub well predicted subset size: 36,
superposes with an RMS of 2.14 A (but for clarity, shown here using MaxSub's threshold of 5.0 A).
|
MaxSub score of 3: T0054AL166_1 |
MaxSub score of 2: T0046AL212_1 |
 |
 |
Model no. 1 from group 166 on CASP3 target T0054.
202 residues in the target.
Superposition of all 116 residues of the model: 8.08 A
MaxSub well predicted subset size: 68,
superposes with an RMS of 2.08 A.
|
Model no. 1 from group 212 on CASP3 target T0046.
119 residues in the target.
Superposition of all 84 residues of the model: 6.62 A
MaxSub well predicted subset size: 34,
superposes with an RMS of 1.53 A.
|
MaxSub score of 1: T0100TS395_1 |
DALI alignment for comparison (MaxSub score of 4): T0100TS000_1 |
 |
 |
Model no. 1 from server 395 on target T0100. Model built on correct parent, but shifted in many of the strands.
Although this model has
54 residues that can be superimposed in a sequence dependent fashion
at a distance of 3.5 A, the score it receives is only 1.3 (or a "1" in the
tables), because this leaves a significant portion of the target
unaligned (288 residues in the target).
Only one other model for T0100 scored above zero with MaxSub and standard parameters.
To
show the difficulty of aligning this target correctly, the DALI alignment is
shown on the right. Server 395 produced the best model for this target in
its model no. 5.
LGscores:
4 and 9.
|
DALI alignment showing the difficulty of aligning this target
correctly, partly because of the presence of a number of loops. The DALI
alignment received a MaxSub of 4 only.
|
MaxSub's False Negative (?) score of 0: T0100TS259_1 |
MaxSub's False Positive (?) scoring model (MaxSub score of 1): T0043TS005_1 |
 |
 |
Model no. 1 from server 259 on target T0100. Model built on correct parent, but shifted in many of the strands, with only 5 to 6 regions
aligned with shift zero. The number of sequence-dependent superimposable
residues at 3.5A falls exactly at MaxSub thresholds. Thus, although this
model might be considered as correct, the automatic evaluation of MaxSub,
based on rigid thresholds, could not award any points to this model.
(see another example of models falling at the cutoff threshold for
target T0121 above).
If credit is deserved for the general topology in a
sequence-independent measure, which disregards
alignment shifts, this model should be considered correct.
In fact, with the
LGscores:
the sequence dependent LGscore assigns a score of 2(0) whereas the
sequence independent LGscore2 assigns a score of 8.
A similar case occurs for a number of other models for targets T0100 and
T0101.
|
Model no.1 from server 005 on CASP3 target T0043.
158 residues in the target.
Superposition of all 158 residues of the model: 14.23 A
MaxSub well predicted subset size: 25,
superposes with RMS of 1.83A.
This is one example detected of a model probably wrongly receiving a score by
MaxSub. The match of 25 residues, and the relatively
small size of the target, lead to a MaxSub score of 0.13, just above its
thresholds. However, it is doubtful that this model deserves any credit.
These two last examples illustrate some of the current limitations of our
automated
sequence dependent evaluation method, and some of the directions required toward
improving it. Thus CAFASP has also been a valuable experiment on
model evaluation techniques.
|