clustalw.ms

经典生物信息学多序列比对工具clustalw

and PILEUP: neither program can generate the correct blocks corresponding to 
the secondary structure elements. 

Figure 4 shows an alignment generated by CLUSTAL W of the example set of 
SH3 domains. The alignment was generated in two steps. After progressive 
alignment, five blocks were produced, corresponding to structural elements, 
with gaps inserted exclusively in the known loop regions. The beta strands in 
blocks 1, 4 and 5 were all correctly superposed. However, four sequences in 
block 2 and one sequence in block 3 were misaligned by 1-2 residues 
(underlined in figure 4). A second progressive alignment of the aligned 
sequences, including the gaps, improved this alignment: A single misaligned 
sequence, H_P55, remains in block 2 (boxed in figure 4), while block 3 is now 
completely aligned.  This alignment corrects several errors (eg. P85A, P85B 
and FUS1) in the manual alignment (23).

The SH3 alignment illustrates several features of CLUSTAL W usage. Firstly, 
in a practical application involving divergent sequences, the initial 
progressive alignment is likely to be a good but not perfect approximation to 
the correct alignment. The alignment quality can be improved in a number of 
ways. If the block structure of the alignment appears to be correct, realignment 
of the alignment will usually improve most of the misaligned blocks: the 
existing gaps allow the blocks to "float" cheaply to a locally optimal position 
without disturbing the rest of the alignment. Remaining sequences which are 
doubtfully aligned can then be individually tested by profile alignment to the 
remainder: the misaligned H_P55 SH3 domain can be correctly aligned by 
profile (with GOP <= 8). The indel regions in the final alignment can then be 
manually cleaned up: Usually the exact alignment in the loop regions is not 
determinable, and may have no meaning in structural terms. It is then 
desirable to have a single gap per structural loop. CLUSTAL W achieved this 
for two of the four SH3 loop regions (figure 4).

If the block structure of the alignment appears suspect, greater intervention by 
the user may be required. The most divergent sequences, especially if they 
have large insertions (which can be discerned with the aid of dot matrix 
plots), should be left out of the progressive alignment. If there are sets of 
closely related sequences that are deeply diverged from other sets, these can be 
separately aligned and then merged by profile alignment. Incorrectly 
determined sequences, containing frameshifts, can also confound regions of 
an alignment: these can be hard to detect but sometimes they have been 
grouped within the excluded divergent sequences: then they may be revealed 
when they are individually compared to the alignment as having apparently 
nonsense segments with respect to the other sequences. 



Finding the best alignment

In cases where all of the sequences in a data set are very similar (e.g. no pair 
less than 35% identical), CLUSTAL W will find an alignment which is 
difficult to improve by eye.  In this sense, the alignment is optimal with 
regard to the alternative of manual alignment.  Mathematically, this is vague 
and can only be put on a more systematic footing by finding an objective 
function (a measure of multiple alignment quality) that exactly mirrors the 
information used by an "expert" to evaluate an alignment.  Nonetheless, if an 
alignment is impossible to improve by eye, then the program has achieved a 
very useful result.   

In more difficult cases, as more divergent sequences are included, it becomes 
increasingly difficult to find good alignments and to evaluate them.    What 
we find with CLUSTAL W is that the basic block-like structure of the 
alignment (corresponding to the major secondary structure elements) is 
usually recovered, with some of the most divergent sequences misaligned in 
small regions.  This is a very useful starting point for manual refinement as it 
helps define the major blocks of similarity.   The problem sequences can be 
removed from the analysis and realigned to the rest of the sequences 
automatically or with different parameter settings.   An examination of the 
tree used to guide the alignment will usually show which sequences will be 
most unreliably placed (those that branch off closest to the root and/or those 
that align to other single sequences at a very low level of sequence identity 
rather than align to a group of pre-aligned sequences).  Finally, one can 
simply iterate the multiple alignment process by feeding an output alignment 
back into CLUSTAL W and repeating the multiple alignment process (using 
the same or different parameters).   The SH3 domain alignment in figure 4 
was derived in this way by 2 passes using default parameters.  In the second 
pass, the local gap penalties are dominated by the placement of the initial 
major gap positions.  The alignment will either remain unchanged or will 
converge rapidly (after 1 or 2 extra passes) on a better solution.  If the 
placement of the initial gaps is approximately correct but some of the 
sequences are locally misaligned, this works well.  


Comparison with other methods

Recently, several papers have addressed the problem of position specific 
parameters for multiple alignment.  In one case (35), local gap penalties are 
increased in alpha helical and beta strand regions, when the 3-D structures of 
one or more of the sequences are known.  In a second case (36), a hidden 
Markov model was used to estimate position specific gap penalties and 
residue substitution weight matrices when large numbers of examples of a 
protein domain were known.  With CLUSTAL W, we attempt to derive the 
same information purely from the set of sequences to be aligned.  Therefore, 
we can apply the method to any set of sequences.  The success of this approach 
will depend on the number of available sequences and their evolutionary 
relationships.  It will also depend on the decision making process during 
multiple alignment (e.g. when to change weight matrix) and the accuracy and 
appropriateness of our parameterisation.  In the long term, this can only be 
evaluated by exhaustive testing of sets of sequences where the correct 
alignment (or parts of it) are known from structural information.   What is 
clear, however, is that the modifications described here significantly improve 
the sensitivity of the progressive multiple alignment approach.  This is 
achieved with almost no sacrifice in speed and efficiency.  

There are several areas where further improvements in sensitivity and 
accuracy can be made.  Firstly, the residue weight matrices and gap settings 
can be made more accurate as more and more data accumulate, while 
matrices for specific sequence types can be derived (e.g. for transmembrane 
regions (37)).  Secondly, stochastic or iterative optimisation methods can be 
used to refine initial alignments (7,9,10).   CLUSTAL W could be run with 
several sets of starting parameters and in each case, the alignments refined 
according to an objective function.   The search for a good objective function, 
that takes into account the sequence and position specific information used in 
CLUSTAL W is a key area of research.   Finally, the average number of 
examples of each protein domain or family is growing steadily.  It is not only 
important that programs can cope with the large volumes of data that are 
being generated, they should be able to exploit the new information to make 
the alignments more and more accurate.   Globally optimal alignments 
(according to an objective function) may not always be possible but the 
problem may be avoided if sufficiently large volumes of data become 
available.  CLUSTAL W is a step in this direction.

ACKNOWLEDGEMENTS

Numerous people have offered advice and suggestions for improvements to 
earlier versions of the CLUSTAL programs.  D.H. wishes to apologise to all of 
the irate CLUSTAL V users who had to live with the bugs and lack of facilities 
for getting trees in the New Hampshire format.  We wish to specifically thank 
Jeroen Coppieters who suggested using a series of weight matrices and Steven 
Henikoff for advice on using the BLOSUM matrices.  We are grateful to Rein 
Aasland, Peer Bork, Ariel Blocker and B巖trand Seraphin for providing 
challenging alignment problems.   T.G. and J.T. thank Kevin Leonard for 
support and encouragement.  Finally, we thank all of the people who were 
involved with various CLUSTAL programs over the years, namely: Paul 
Sharp, Rainer Fuchs and Alan Bleasby.


REFERENCES

 1.Feng, D.-F. and Doolittle, R.F. (1987). J. Mol. Evol. 25, 351-360.
 2.Needleman, S.B. and Wunsch, C.D. (1970). J. Mol. Biol. 48, 443-453.
 3.Dayhoff, M.O., Schwartz, R.M. and Orcutt, B.C. (1978)  in Atlas of Protein 
Sequence and Structure, vol. 5, suppl. 3 (Dayhoff, M.O., ed.), pp 345-352, 
NBRF, Washington.
 4.Henikoff, S. and Henikoff, J.G. (1992). Proc. Natl. Acad. Sci. USA 89, 10915-
10919.
 5.Lipman, D.J., Altschul, S.F. and Kececioglu, J.D. (1989). Proc. Natl. Acad. Sci. 
USA 86, 4412-4415.
 6.Barton, G.J. and Sternberg, M.J.E. (1987). J. Mol. Biol. 198, 327-337.
 7.Gotoh, O. (1993). CABIOS 9, 361-370.
 8.Altschul, S.F. (1989). J. Theor. Biol. 138, 297-309.
 9.Lukashin, A.V., Engelbrecht, J. and Brunak, S. (1992). Nucl. Acids Res. 20, 
2511-2516.
10.Lawrence, C.E., Altschul, S.F., Boguski, M.S., Liu, J.S., Neuwald, A.F. and 
Wooton, J.C. (1993). Science, 262, 208-214.
11.Vingron, M. and Waterman, M.S. (1993). J. Mol. Biol. 234, 1-12.
12.Pascarella, S. and Argos, P. (1992). J. Mol. Biol. 224, 461-471.
13.Collins, J.F. and Coulson, A.F.W. (1987). In Nucleic acid and protein 
sequence analysis a practical approach, Bishop, M.J. and Rawlings, C.J. ed., 
chapter 13, pp. 323-358.
14.Vingron, M. and Sibbald, P.R. (1993). Proc. Natl. Acad. Sci. USA, 90, 8777-
8781.
15.Thompson, J.D., Higgins, D.G. and Gibson, T.J. (1994). CABIOS, 10, 19-29.
16.L焧hy, R., Xenarios, I. and Bucher, P. (1994). Protein Science, 3, 139-146.
17.Higgins, D.G. and Sharp, P.M. (1988). Gene, 73, 237-244.
18.Higgins, D.G. and Sharp, P.M. (1989). CABIOS, 5, 151-153.
19.Higgins, D.G., Bleasby, A.J. and Fuchs, R. (1992). CABIOS, 8, 189-191.
20.Sneath, P.H.A. and Sokal, R.R. (1973). Numerical Taxonomy, W.H. 
Freeman, San Francisco.
21.Saitou, N. and Nei, M. (1987). Mol. Biol. Evol. 4, 406-425.
22.Wilbur, W.J. and Lipman, D.J. (1983). Proc. Natl. Acad. Sci. USA, 80, 726-
730.
23.Musacchio, A., Gibson, T., Lehto, V.-P. and Saraste, M. (1992). FEBS Lett. 
307, 55-61.
24.Musacchio, A., Noble, M., Pauptit, R., Wierenga, R. and Saraste, M. (1992). 
Nature, 359, 851-855.
25.Bashford, D., Chothia, C. and Lesk, A.M. (1987). J. Mol. Biol. 196, 199-216.
26.Myers, E.W. and Miller, W. (1988). CABIOS, 4, 11-17.
27.Thompson, J.D. (1994). CABIOS, (Submitted).
28.Smith, T.F., Waterman, M.S. and Fitch, W.M. (1981). J. Mol. Evol. 18, 38-46.
29.Pearson, W.R. and Lipman, D.J. (1988). Proc. Natl. Acad. Sci. USA. 85, 2444-
2448.
30.Devereux, J., Haeberli, P. and Smithies, O. (1984). Nucleic Acids Res. 12, 
387-395.
31.Felsenstein, J. (1989). Cladistics 5, 164-166.
32.Kimura, M. (1980). J. Mol. Evol. 16, 111-120.
33.Kimura, M. (1983). The Neutral Theory of Molecular Evolution.  
Cambridge University Press, Cambridge.
34.Felsenstein, J. (1985). Evolution 39, 783-791.
35.Smith, R.F. and Smith, T.F. (1992) Protein Engineering 5, 35-41.
36.Krogh, A., Brown, M., Mian, S., Sj歭ander, K. and Haussler, D. (1994) J. Mol. 
Biol. 235-1501-1531.
37.Jones, D.T., Taylor, W.R. and Thornton, J.M.  (1994). FEBS Lett. 339, 269-275.
38.Bairoch, A. and B歝kmann, B. (1992) Nucleic Acids Res., 20, 2019-2022.
39.Noble, M.E.M., Musacchio, A., Saraste, M., Courtneidge, S.A. and 
Wierenga, R.K. (1993) EMBO J. 12, 2617-2624.
40.Kabsch, W. and Sander, C. (1983) Biopolymers, 22, 2577-2637.

FIGURE LEGENDS

Figure 1.  The basic progressive alignment procedure, illustrated using a set of 
7 globins of known tertiary structure.  The sequence names are from Swiss 
Prot (38):  Hba_Horse: horse alpha globin; Hba_Human: human alpha globin; 
Hbb_Horse: horse beta globin; Hbb_Human: human beta globin; Myg_Phyca: 
sperm whale myoglobin; Glb5_Petma: lamprey cyanohaemoglobin; 
Lgb2_Luplu: lupin leghaemoglobin.   In the distance matrix, the mean 
number of differences per residue is given.  The unrooted tree shows all 
branch lengths drawn to scale.  In the rooted tree, all branch lengths (mean 
number of differences per residue along each branch) are given as well as 
weights for each sequence.  In the multiple alignment, the approximate 
positions of the 7 alpha helices, common to all 7 proteins are shown.  This 
alignment was derived using CLUSTAL W with default parameters and the 
PAM (3) series of weight matrices.  

Figure 2.  The scoring scheme for comparing two positions from two 
alignments.   Two sections of alignment with 4 and 2 sequences respectively 
are shown.   The score of the position with amino acids T,L,K,K versus the 
position with amino acids V and I is given with and without sequence 
weights.  M(X,Y) is the weight matrix entry for amino acid X versus amino 
acid Y.  Wn is the weight for sequence n.

Figure 3.  The variation in local gap opening penalty is plotted for a section of 
alignment.  The inital gap opening penalty is indicated by a dotted line. Two 
hydrophilic stretches are underlined.  The lowest penalties correspond to the 
ends of the alignment, the hydrophilic stretches and the two positions with 
gaps.   The highest values are within 8 residues of the two gap positions.  The 
rest of the variation is caused by the residue specific gap penalties (12).

Figure 4.  CLUSTAL W Alignment of a set of SH3 domains taken from (23). 
Secondary structure assignments for the solved Spectrin (24) and Fyn (39) 
domains are according to DSSP (40). The alignment was generated in two 
steps using default parameters. After full multiple alignment, the aligned 
sequences were realigned. Segments which were correctly aligned in the 
second pass are underlined. The single misaligned segment in H_P55 and the 
misaligned residue in H_NCK/2 are boxed.

The sequences are coloured to illustrate significant features. All G (orange) 
and P (yellow) are coloured. Other residues matching a frequent occurrence of 
a property in a column are coloured: hydrophobic = blue; hydrophobic 
tendency = light blue; basic = red; acidic = purple; hydrophilic = green; White 
= unconserved. The alignment figure was prepared with the GDE sequence 
editor (S. Smith, Harvard University) and COLORMASK (J. Thompson, 
EMBL).




Table 1.  Pascarella and Argos residue specific gap modification factors.   
-----------------------------------------------------------------------------------
A	1.13		M	1.29
C	1.13		N	0.63
D	0.96		P	0.74
E	1.31		Q	1.07
F	1.20		R	0.72
G	0.61		S	0.76
H	1.00		T	0.89
I	1.32		V	1.25
K	0.96		Y	1.00
L	1.21		W	1.23
-----------------------------------------------------------------------------------
The values are normalised around a mean value of 1.0 for H.  The lower the 
value, the greater the chance of having an adjacent gap.  These are derived 
from the original table of relative frequencies of gaps adjacent to each residue 
(12) by subtraction from 2.0.