1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
|
******************************************************************************
CLUSTALW-MPI: ClustalW Analysis Using Grid and Parallel Computing
based on ClustalW, the multiple sequence alignment program
(version 1.82, Feb 2001)
Please send bug reports, comments etc. to
kuobin@bii.a-star.edu.sg (Kuo-Bin Li)
******************************************************************************
COPYRIGHT and LICENSING POLICY
CLUSTALW-MPI is freely available to the user community. You can
redistribute it and/or modify it. Since CLUSTALW-MPI was derived from
CLUSTAL W, the original license policy of CLUSTAL W is listed here:
Clustal W is freely available to the user community. However, Clustal W is
increasingly being distributed as part of commercial sequence analysis
packages. To help us safeguard future maintenance and development, commercial
distributors of Clustal W must take out a NON-EXCLUSIVE LICENCE. Anyone
wishing to commercially distribute version 1.81 of Clustal W should contact the
authors unless they have previously taken out a licence.
******************************************************************************
ClustalW is a popular tool for multiple sequence alignment. The
alignment is achieved via three steps: pairwise alignment,
guide-tree generation and progressive alignment. ClustalW-MPI is an
MPI implementation of ClustalW. Based on
version 1.82 of the original ClustalW, both the pairwise
and progressive alignments are parallelized with MPI, a
popular message passing programming standard. The
pairwise alignments can be easily parallelized since the many
alignments are time independent on each other. However
the progressive alignments are essentially not parallelizable
because of the time dependencies between each alignment.
Here we applied the recursive parallelism paradigm to the linear space
profile-profile alignment algorithm. This approach is more time
efficient on computers with distributed memory architecture.
Traditional approach that relies on precomputing the profile-profile
score matrix has also been implemented. Results shown the latter is indeed
more appropriate for shared memory multiprocessor computer.
The software is available at
http://www.bii.a-star.edu.sg/~kuobin/clustalw-mpi/
The original ClustalW/ClustalX can be found at ftp://ftp-igbmc.u-strasbg.fr.
REFERENCE
---------
Kuo-Bin Li (2003) "ClustalW-MPI: ClustalW Analysis Using Distributed and Parallel Computing",
Bioinformatics, in press.
INSTALLATION (for Unix/Linux)
------------
This is an extremely quick installation guide.
1. Make sure you have MPICH or LAM installed on your system.
2. Unpack the package in any working directory:
tar xvfp clustalw-mpi-0.1.tar.gz
3. Take a look at the Makefile and make the modifications that you might desire,
in particular:
CC = mpicc
CFLAGS = -c -g
or
CFLAGS = -c -O3
4. Build the whole thing simply by typing "make".
5. If you wanted to use serial codes to compute the neighbor-joining tree,
you would have to define the macro "SERIAL_NJTREE" when compiling trees.c:
CFLAGS = -c -g -DSERIAL_NJTREE
This macro is defined in the default Makefile. That is, to use
MPI codes in neighbor-joining tree, you have to "undefine" the
macro "SERIAL_NJTREE" in your Makefile.
SAMPLE USAGE (for Unix/Linux)
------------
1. To make a full multiple sequence alignment:
(using one master node and 4 computing nodes)
%mpirun -np 5 ./clustalw-mpi -infile=dele.input
%mpirun -np 5 ./clustalw-mpi -infile=CFTR.input
2. To make a guide tree only:
%mpirun -np 5 ./clustalw-mpi -infile=dele.input -newtree=dele.mytree
%mpirun -np 5 ./clustalw-mpi -infile=CFTR.input -newtree=CFTR.mytree
3. To make a multiple sequence alignment out of an existing
tree:
%mpirun -np 5 ./clustalw-mpi -infile=dele.input -usetree=dele.mytree
%mpirun -np 5 ./clustalw-mpi -infile=CFTR.input -usetree=CFTR.mytree
4. The environment variable, CLUSTALG_PARALLEL_PDIFF, could be used to
run the progressive alignment based on the parallelized pdiff().
By default the variable CLUSTALG_PARALLEL_PDIFF is not set, and
the progressive alignment will be parallelized accroding the structure
of the neighbor-joining tree. However, parallelized pdiff() will still
be used in the later stage when prfalign() tries to align more distant
sequences to the profiles. If you don't understand this,
simply leave the variable unset.
KNOWN PROBLEM
------------
1. On Intel IA32 platforms, slightly different neighbor-joining trees
might be obtained with and without enabling the compiler's optimization flags.
This is due to the fact that Intel processors use 80-bit FPU registers
to cache "double" variables, which are supposed to be 64-bit long. With '-O1'
or above optimizer flag, the compiler would not always immediately save the
variables involved in a double operation back to memory. Instead, intermediate
results will be saved in registers, having 80-bit of precision. This would
cause problem for nj_tree() because it is sensitive to the precision of floating
point numbers.
Solutions:
(1) Other platforms, including Intel's IA64, don't seem to have this problem.
or
(2) Building "trees.c" with optios like the below: (potentially
with high performance overhead)
%gcc -c -O3 -ffloat-store trees.c // GNU gcc
%icc -c -O3 -mp trees.c // Intel C compiler
or
(3) Decalring relevant variables as "volatile" in nj_tree():
volatile double diq, djq, dij, d2r, dr, dio, djo, da;
volatile double *rdiq;
rdiq = (volatile double *)malloc(((last_seq-first_seq+1)+1)*
sizeof(volatile double));
...
...
free((void*)rdiq);
ACKNOWLEDGEMENTS
------------
1. In parallel_compare.c: "fprintf(stderr, ..." changed to "fprintf(stdout,....".
Thanks to Ville Silventoinen <vsi@ebi.ac.uk>.
|
