advertisement: compare things at compare-stuff.com!
next up previous contents
Next: Sequence Conservation Up: Results Previous: Baseline comparison   Contents

Alignments of hydrophobicity-related information

In this section we use the SIVA method to align hydrophobicity profiles, i.e. the property vector $P_i$ contains as its single component the mean hydrophobicity, $\overline{h}$, calculated at sequence position $i$ in the multiple sequence alignment. The summary fold recognition results, including $\overline{R}_{adj}$, $\overline{S}$ and $T$, are presented in Table 4.4 for the four hydrophobicity scales tested (shown in Table 4.3). The Kyte and Doolittle hydrophobicity scale achieves the highest number of correct top-ranking fold recognitions ($T=8$); a little better than the Smith Waterman results. Amongst the trials using hydrophobicity related information, the Eisenberg and McLachlan direction of hydrophobic moment measure gives the best overall ranking ( $\overline{R}_{adj}=10.8$) and alignments ( $\overline{S}=20.9$). The performance measures vary considerably within the three fold topologies also shown in Table 4.4. Each of the four measures performs best in one or other category. Most notably the hydrophobic moment direction gives the best ranking for the immunoglobulin-like folds (2.60.40), but gives undistinguished alignments.


Table 4.3: Four hydrophobicity related amino acid indices.
amino acid Kyte & Bull & Radzicka & Eisenberg &
  Doolittle1 Breese2 Wolfenden3 McLachlan4
I 4.5 -2.26 1.31 0.99
V 4.2 -1.56 1.09 0.84
L 3.8 -2.46 1.21 0.89
F 2.8 -2.33 1.27 0.92
C 2.5 -0.45 1.36 0.76
M 1.9 -1.47 1.27 0.94
A 1.8 -0.20 -0.06 0 (n/a)5
G -0.4 0.00 -0.41 0 (n/a)
T -0.7 -0.52 -0.27 0.09
S -0.8 -0.39 -0.50 -0.67
W -0.9 -2.01 0.88 0.67
Y -1.3 -2.24 0.33 -0.93
P -1.6 -0.98 0.00 0.22
H -3.2 -0.12 0.49 -0.75
Q -3.5 0.16 -0.73 -1.00
N -3.5 0.08 -0.48 -0.86
E -3.5 -0.30 -0.77 -0.89
D -3.5 -0.20 -0.80 -0.98
K -3.9 -0.35 -1.18 -0.99
R -4.5 -0.12 -0.84 -0.96
1 [Kyte & Doolittle, 1982]
2 [Bull & Breese, 1974]
3 [Radzicka & Wolfenden, 1988]
4 [Eisenberg & McLachlan, 1986]
5  Too few side-chain atoms to calculate moment





Table 4.4: Summary of fold recognition using alignment of raw sequence derived information.
Sequence whole library 3.40.3301 3.20.40 2.60.40
information $T$ $\overline{R}_{adj}$ $\overline{S}$ $\overline{R}_{adj}$ $\overline{S}$ $\overline{R}_{adj}$ $\overline{S}$ $\overline{R}_{adj}$ $\overline{S}$
Kyte & Doolittle 8 11.4 22.4 6.7 18.1 5.4 26.0 12.9 30.1
                   
Radzicka & 6 12.0 22.6 5.1 17.3 6.2 32.5 21.2 28.5
Wolfenden                  
Bull & Breese 7 14.2 23.2 5.6 20.5 7.8 29.7 16.0 27.1
                   
Eisenberg & 6 10.8 20.9 7.0 16.8 5.7 29.8 9.2 29.2
McLachlan                  
Conservation 2 17.0 25.1 10.9 22.8 9.7 31.3 11.7 19.3
                   
Conserved 9 10.8 23.2 5.7 19.9 6.2 27.1 17.9 29.5
Hydrophobicity                  
Hydrophobicity & 10 10.5 23.6 7.5 21.2 4.6 22.5 13.4 30.3
Conservation 20:1                  
DSC prediction 8 9.0 26.8 5.2 22.3 5.3 33.3 3.7 25.9
                   
Hydrophobicity & 10 9.5 21.5 5.4 17.6 4.1 23.4 6.3 29.0
DSC prediction 2:1                  
Hydrophobicity & 13 7.7 22.2 4.4 18.0 3.6 23.9 2.2 29.1
DSC prediction 1:1                  
Hydrophobicity & 10 7.6 24.7 5.2 20.9 4.1 30.8 2.0 28.2
DSC prediction 1:2                  
1 see Appendix B for descriptions of CATH codes





Table 4.5: Summary of fold recognition results using alignments of hydrophobicity (Kyte and Doolittle). Only `correct' query-library pairs are shown.
rank by   domain CATH length  
all1 query2 Z-score query library topology query library $\overline{S}$3
1 1 1.940 5p2100 1hurA0 3.40.330 166 180 1.0
2 1 1.860 1hurA0 5p2100 3.40.330 180 166 0.9
5 1 0.904 1pii01 1tpfA0 3.20.40 261 250 22.0
15 1 0.775 1tpfA0 1pii01 3.20.40 250 261 25.4
17 1 0.758 1dgd02 1aam02 3.40.640 264 271 17.8
26 1 0.696 1llo00 1pii01 3.20.40 273 261 24.4
28 3 0.684 1atnA2 1atr03 3.30.420 108 107 4.2
32 2 0.673 1atr03 1atnA2 3.30.420 107 108 3.9
35 1 0.655 1ntr00 4fxn00 3.40.330 124 138 5.0
47 3 0.608 1llo00 1tpfA0 3.20.40 273 250 9.4
48 1 0.603 4fxn00 1ntr00 3.40.330 138 124 3.7
1 global ranking (out of 2214 query-library pairs) by Z-score calculated per query
2 ranked by alignment score for each query (out of 82 pairwise alignments) -- therefore a `1' in this column indicates a correct top ranking alignment.
3 mean alignment shift





Table 4.6: Summary fold recognition results using alignments of hydrophobicity (Bull and Breese). Only `correct' query-library pairs are shown.
rank by   domain CATH length  
all query Z-score query library topology query library $\overline{S}$
1 1 2.466 1hurA0 5p2100 3.40.330 180 166 0.8
2 1 2.335 5p2100 1hurA0 3.40.330 166 180 0.7
7 3 0.982 1atr03 1atnA2 3.30.420 107 108 2.6
11 1 0.828 1atnA2 1atr03 3.30.420 108 107 4.3
19 3 0.761 1cnd01 1pkm03 2.40.90 106 103 18.3
27 3 0.724 1llo00 1nal10 3.20.40 273 291 21.8
33 1 0.701 1pkm03 1cnd01 2.40.90 103 106 14.1
41 4 0.674 1dgd02 1aam02 3.40.640 264 271 19.4
49 5 0.642 1nal10 1llo00 3.20.40 291 273 21.0
51 1 0.641 1ntr00 5p2100 3.40.330 124 166 3.9
58 1 0.613 4fxn00 1ntr00 3.40.330 138 124 2.0
60 8 0.607 1aam02 1dgd02 3.40.640 271 264 23.2
64 5 0.600 1llo00 1tpfA0 3.20.40 273 250 10.2
71 7 0.584 1llo00 1pii01 3.20.40 273 261 37.2
72 2 0.574 1ntr00 4fxn00 3.40.330 124 138 2.5
73 1 0.569 2ohxA2 1pnrA3 3.40.330 139 147 8.9





Table 4.7: Summary of fold recognition results using alignments of hydrophobic moment direction (Eisenberg and McLachlan). Only `correct' query-library pairs are shown.
rank by   domain CATH length  
all query Z-score query library topology query library $\overline{S}$
1 1 2.085 5p2100 1hurA0 3.40.330 166 180 0.6
2 1 1.980 1hurA0 5p2100 3.40.330 180 166 0.5
8 1 0.813 1dgd02 1aam02 3.40.640 264 271 15.7
10 1 0.781 1ntr00 4fxn00 3.40.330 124 138 4.0
16 1 0.718 1pii01 1tpfA0 3.20.40 261 250 22.2
33 4 0.643 1pii01 1nal10 3.20.40 261 291 9.3
34 1 0.640 4fxn00 1ntr00 3.40.330 138 124 2.8
38 2 0.623 1llo00 1nal10 3.20.40 273 291 21.8
44 2 0.606 1nal10 1pii01 3.20.40 291 261 8.9
45 3 0.602 1nal10 1tpfA0 3.20.40 291 250 17.1

More detailed results for Kyte and Doolittle hydrophobicity, Bull and Breese hydrophobicity, and the hydrophobic moment direction are given in Tables 4.5, 4.6 and 4.7. These tables show all the correct top-ranking fold predictions, and other correct pairings. The tables are sorted by Z-score (these are calculated separately for each query). From these tables one can judge what proportion of pairings are correct above a certain Z-score threshold. In these experiments (and almost all that follow), the two domains 5p2100 and 1hurA0 align with each other to give the highest ranking Z-scores. These domains were also detected by the Smith Waterman method and are clearly the `easiest' pair to recognise, however the other correct top ranking pairs are not always the same as the Smith Waterman hits (in Table 4.2). Note also that the Smith Waterman algorithm has identified similarities between domains of very different length (for example 1tpfA0 and 1pii02) through the use of a local alignment algorithm. In contrast the SIVA top hits all have much smaller differences in length.

next up previous contents
Next: Sequence Conservation Up: Results Previous: Baseline comparison   Contents
Copyright Bob MacCallum - DISCLAIMER: this was written in 1997 and may contain out-of-date information.