SSM vs. others: 1TIG

 
Materials from this page cannot be reproduced without permission from the authors.
Comparisons made on November 2002, using current versions of VAST, CE, DALI, DEJAVU and SSM v1.22 from 20/11/2002.
 

CONTENTS
  1. VAST
  2. CE (Combinatorial Extension)
  3. DALI
  4. DEJAVU
  5. Conclusion
 
1TIG (88 residues)
TRANSLATION INITIATION FACTOR 3 C-TERMINAL DOMAIN
Two helices and 4 strands were used for SSE matching.
 

1.  V A S T    (server)

Figure 1TIG-1 shows the Ca-alignment lengths obtained from SSM and VAST for different structural neighbours (as chosen by VAST). As seen from the picture, SSM offers longer alignments, with the exception for the most remote structures. The general trend of both SSM and VAST curves look similar (taking into account that PDB entries are ordered by decreasing SSM alignment length, which makes the SSM curve smooth), however there is a clear non-monotonicity in the comparison: two alignments that have approximately the same length in SSM may have a very different alignment lengths in VAST and vice versa.


  Figure 1TIG-1.
Length of Ca-alignment as a function of PDB entry, obtained by SSM (black line) and VAST (red line). Details of the calculations are given here.
 

As shown in Figure 1TIG-2, longer SSM alignments have higher RMSDs, which is an indication of that SSM and VAST employ different criteria for solving the compromise between alignment length and RMSD.


  Figure 1TIG-2.
RMSD of Ca-alignment corresponding to data in Figure 1TIG-1. Details of the calculations are given here.
 

As Figure 1TIG:A-3 shows, SSM and VAST agree well on the match index of Ca-alignment. This proves that both servers provide the same principal quality of alignment.


  Figure 1TIG-3.
Match Index corresponding to data shown in Figure 1SAR:A-1. Details of the calculations are given here.
 

As seen from Figure 1TIG-4, SSM gives P-values varying in a wider range than those obtained from VAST. SSM's P-values are on average lower and look more scattered.


  Figure 1TIG-4.
P-values corresponding to matches shown in Figure 1TIG-1. Details of the calculations are given here.
 

Z-scores of SSM's Ca-alignments (Figure 1TIG-5) show better, than P-values, agreement with VAST results. They span over approximately the same range and show a weak trend to decline with growing structural dissimilarity.


  Figure 1TIG-5.
Z-scores corresponding to matches shown in Figure 1TIG-1. Details of the calculations are given here.
 

 

 

2.  C E (Combinatorial Extension)    (server)

Comparison with Combinatorial Extension shows that, as a rule, SSM gives shorter Ca-alignments (cf. Figure 1TIG-6). The difference between the servers grows with growing structural dissimilarity at very high non-monotonicity. The latter means that two matches having same alignment length in SSM results, receive very different alignment lengths in CE output (and, correspondingly, vice versa). As may be seen from Figure 1TIG-6, the non-monotonicity reaches 20 residues, which is nearly 25% of the input sequence length.


  Figure 1TIG-6.
Length of Ca-alignment as a function of PDB entry, obtained by SSM (black line) and CE (red line). Details of the calculations are given here.
 

Figure 1TIG-7 shows that CE produces considerably higher RMSDs, as compared to SSM. Many of CE alignments have RMSD exceeding 5 Å. Just as in the case of alignment lengths (Figure 1TIG-6), comparison of the results is very non-monotonic.


  Figure 1TIG-7.
RMSD of Ca-alignment corresponding to data in Figure 1TIG-6. Details of the calculations are given here.
 

The match index of SSM and CE alignments, shown in Figure 1TIG-8, demonstrates a much better agreement than alignment lengths and RMSDs separately. This indicates that SSM and CE perform alignments on approximately the same quality level, with different balancing the alignment length and RMSD. However, in this particular case, the difference between SSM and CE match index is more than usual.


  Figure 1TIG:A-8.
Match Index corresponding to data shown in Figure 1TIG-6. Details of the calculations are given here.
 

Z-scores of both SSM and CE alignments do not show any specific trends at given ordering of the results (this implies, no clear correlation between Z-score and alignment length), see Figure 1TIG-9. SSM results look more scattered than those of CE. CE's Z-scores are roughly twice lower than SSM's ones.


  Figure 1TIG-9.
Z-scores corresponding to matches shown in Figure 1TIG-6. Details of the calculations are given here.
 

 

 

3.  D A L I    (server)

Ca-alignments from DALI do not match those from SSM precisely (see Figure 1TIG-10), however their difference is relatively well centered (meaning that there are approximately as many positive deviations as the negative ones). Generally the comparison is very non-monotonic, similar to what is observed for VAST and CE (cf. Figures 1TIG-1 and 1TIG-6).


  Figure 1TIG-10.
Length of Ca-alignment as a function of PDB entry, obtained by SSM (black line) and DALI (red line). Details of the calculations are given here.
 

Similarly to alignment lengths, the alignment RMSDs from SSM and DALI are widely scattered, however they cover approximately the same range and look well centered respecting to each other (see Figure 1TIG-11). A detail analysis of Figures 1TIG-10 and 1TIG-11 reveals that in fact they are reasonably well correlated, i.e. positive/negative deviations in Figure 1TIG-10 correspond to positive/negative deviations in Figure 1TIG-11.


  Figure 1TIG-11.
RMSD of Ca-alignment corresponding to data in Figure 1TIG-10. Details of the calculations are given here.
 

The above conclusion is well corroborated by data presented in Figure 1TIG-12, where the correlation between Figures 1TIG-10 and 1TIG-11 is expressed in close values of SSM and DALI match indexes. As may be seen from the Figure, despite non-monotonicity of both SSM and DALI curves, their actual deviation is not large. This is an indication of the fact that SSM and DALI produce 3D alignments of approximately equal quality, while analysis of Figures 1TIG-10 and 1TIG-11 suggests that alignment length and RMSD are balanced differently.


  Figure 1TIG-12.
Match Index corresponding to data shown in Figure 1TIG-10. Details of the calculations are given here.
 

As in most examples, DALI's Z-scores are higher than those from SSM for highly similar structures, and lower for the remote ones (Figure 1TIG-13). Typically as well, Z-scores from DALI agree better with minus logarithm of SSM's P-values (shown by black line in Figure 1TIG-13).


  Figure 1TIG-13.
Z-scores corresponding to matches shown in Figure 1TIG-10. Details of the calculations are given here.
 

 

 

4.  D E J A V U    (server)

DEJAVU failed to recognize most closest structural neighbours, including 1TIG itself and gave 1XYZ as the closest prototype of the input. As seen from Figure 1TIG-14 almost all DEJAVU Caalignments are considerably shorter than those from SSM. As appears, the observed difference does not allow for speaking about any similarity.


  Figure 1TIG-14.
Length of Ca-alignment as a function of PDB entry, obtained by SSM (black line) and DEJAVU (red line). Details of the calculations are given here.
 

Expectably, RMSD of SSM alignments are considerably higher than those obtained from DEJAVU, although stay in a fairly reasonable range (less than 5 Å, cf. Figure 1TIG-15). As appears from the Figure, DEJAVU is tuned to keep RMSD below than 2 Å.


  Figure 1TIG-15.
RMSD of Ca-alignment corresponding to data in Figure 1TIG-14. Details of the calculations are given here.
 

The match index shows a somewhat better agreement than alignment lengths and RMSDs (cf. Figure 1COL:A-16). For both SSM and DEJAVU results, match index is scattered over approximately the same range of values, indicating that alignment lengths and RMSDs are correlated in respect to approximately equal quality of match. Nevertheless, the overall difference between SSM and DEJAVU appears to be larger than normally observed.


  Figure 1TIG-16.
Match Index corresponding to data shown in Figure 1TIG-14. Details of the calculations are given here.
 

P-values of SSM and DEJAVU alignments, shown in Figure 1TIG-17, hardly show any trend or similarity to discuss. Generally P-values from DEJAVU are lower, which means that, comparing to SSM, DEJAVU considers its hits as more significant. The general look of the picture suggests, once again, that DEJAVU selects mostly remote structural neighbours, which do not have much statistical significance.


  Figure 1TIG-17.
P-values corresponding to matches shown in Figure 1TIG-14. Details of the calculations are given here.
 

Z-scores, presented in Figure 1TIG-18, show the same type of behaviour as P-values (cf. Figure 1TIG-17) and do not allow to detect a reasonable agreement or correlation. There is a correspondence between Figures 1TIG-17 and 1TIG-18, in particular, Z-scores from DEJAVU are noticeably higher.


  Figure 1TIG-18.
Z-scores corresponding to matches shown in Figure 1TIG-14. Details of the calculations are given here.
 

 

 

5.  Conclusion

In this particular example of PDB 1TIG, SSM, VAST, CE, DALI and DEJAVU demonstrate a somewhat poorer, than normally, agreement in principal quality of 3D Ca-alignments, measured by match index. Nevertheless, it is possible to rate the agreement of match indexes as reasonably good. This is particularly true for DALI, which compares with SSM better than others. A general agreement in match index is not reflected in alignment lengths and RMSDs, which do not compare well.

VAST produces shorter alignments than SSM at lower RMSDs. The comparison is very non-monotonic: same-quality matches may have as much as 40% difference in alignment lengths and RMSDs. The same is true for CE and DALI, however CE gives longer alignments at higher RMSDs, and DALI's results are quite well centered in respect to those of SSM. Comparison with DEJAVU can hardly reveal any similarity (except that in match indexes); as always, DEJAVU gives quite short alignments at RMSD restricted by 2 Å or so.

P-values and Z-scores appear to be less indicative in all results. Their comparison also show high non-monotonicity, and only very basic trends may be derived. These do not make a solid ground for conclusions.