line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
# |
2
|
|
|
|
|
|
|
# BioPerl module for Bio::Restriction::IO::base |
3
|
|
|
|
|
|
|
# |
4
|
|
|
|
|
|
|
# Please direct questions and support issues to |
5
|
|
|
|
|
|
|
# |
6
|
|
|
|
|
|
|
# Cared for by Rob Edwards |
7
|
|
|
|
|
|
|
# |
8
|
|
|
|
|
|
|
# Copyright Rob Edwards |
9
|
|
|
|
|
|
|
# |
10
|
|
|
|
|
|
|
# You may distribute this module under the same terms as perl itself |
11
|
|
|
|
|
|
|
# |
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
# POD documentation - main docs before the code |
14
|
|
|
|
|
|
|
|
15
|
|
|
|
|
|
|
=head1 NAME |
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
Bio::Restriction::IO::base - base enzyme set |
18
|
|
|
|
|
|
|
|
19
|
|
|
|
|
|
|
=head1 SYNOPSIS |
20
|
|
|
|
|
|
|
|
21
|
|
|
|
|
|
|
Do not use this module directly. Use it via the Bio::Restriction::IO class. |
22
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
=head1 DESCRIPTION |
24
|
|
|
|
|
|
|
|
25
|
|
|
|
|
|
|
|
26
|
|
|
|
|
|
|
This class defines some base methods for restriction enzyme input and |
27
|
|
|
|
|
|
|
at the same time gives a base list of common enzymes. |
28
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
=head1 FEEDBACK |
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
=head2 Mailing Lists |
32
|
|
|
|
|
|
|
|
33
|
|
|
|
|
|
|
User feedback is an integral part of the evolution of this and other |
34
|
|
|
|
|
|
|
Bioperl modules. Send your comments and suggestions preferably to the |
35
|
|
|
|
|
|
|
Bioperl mailing lists Your participation is much appreciated. |
36
|
|
|
|
|
|
|
|
37
|
|
|
|
|
|
|
bioperl-l@bioperl.org - General discussion |
38
|
|
|
|
|
|
|
http://bioperl.org/wiki/Mailing_lists - About the mailing lists |
39
|
|
|
|
|
|
|
|
40
|
|
|
|
|
|
|
=head2 Support |
41
|
|
|
|
|
|
|
|
42
|
|
|
|
|
|
|
Please direct usage questions or support issues to the mailing list: |
43
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
I |
45
|
|
|
|
|
|
|
|
46
|
|
|
|
|
|
|
rather than to the module maintainer directly. Many experienced and |
47
|
|
|
|
|
|
|
reponsive experts will be able look at the problem and quickly |
48
|
|
|
|
|
|
|
address it. Please include a thorough description of the problem |
49
|
|
|
|
|
|
|
with code and data examples if at all possible. |
50
|
|
|
|
|
|
|
|
51
|
|
|
|
|
|
|
=head2 Reporting Bugs |
52
|
|
|
|
|
|
|
|
53
|
|
|
|
|
|
|
Report bugs to the Bioperl bug tracking system to help us keep track |
54
|
|
|
|
|
|
|
the bugs and their resolution. Bug reports can be submitted via the |
55
|
|
|
|
|
|
|
web: |
56
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
https://github.com/bioperl/bioperl-live/issues |
58
|
|
|
|
|
|
|
|
59
|
|
|
|
|
|
|
=head1 AUTHOR |
60
|
|
|
|
|
|
|
|
61
|
|
|
|
|
|
|
Rob Edwards, redwards@utmem.edu |
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
=head1 CONTRIBUTORS |
64
|
|
|
|
|
|
|
|
65
|
|
|
|
|
|
|
Heikki Lehvaslaiho, heikki-at-bioperl-dot-org |
66
|
|
|
|
|
|
|
Mark A. Jensen, maj-at-fortinbras-dot-us |
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
=head1 APPENDIX |
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
The rest of the documentation details each of the object |
71
|
|
|
|
|
|
|
methods. Internal methods are usually preceded with a _ |
72
|
|
|
|
|
|
|
|
73
|
|
|
|
|
|
|
=cut |
74
|
|
|
|
|
|
|
|
75
|
|
|
|
|
|
|
# Let the code begin... |
76
|
|
|
|
|
|
|
|
77
|
|
|
|
|
|
|
package Bio::Restriction::IO::base; |
78
|
|
|
|
|
|
|
|
79
|
4
|
|
|
4
|
|
15
|
use strict; |
|
4
|
|
|
|
|
5
|
|
|
4
|
|
|
|
|
101
|
|
80
|
|
|
|
|
|
|
|
81
|
4
|
|
|
4
|
|
398
|
use Bio::Restriction::Enzyme; |
|
4
|
|
|
|
|
6
|
|
|
4
|
|
|
|
|
88
|
|
82
|
4
|
|
|
4
|
|
310
|
use Bio::Restriction::EnzymeCollection; |
|
4
|
|
|
|
|
3
|
|
|
4
|
|
|
|
|
65
|
|
83
|
4
|
|
|
4
|
|
892
|
use Bio::Restriction::Enzyme::MultiCut; |
|
4
|
|
|
|
|
6
|
|
|
4
|
|
|
|
|
82
|
|
84
|
4
|
|
|
4
|
|
824
|
use Bio::Restriction::Enzyme::MultiSite; |
|
4
|
|
|
|
|
5
|
|
|
4
|
|
|
|
|
92
|
|
85
|
|
|
|
|
|
|
|
86
|
4
|
|
|
4
|
|
21
|
use base qw(Bio::Restriction::IO); |
|
4
|
|
|
|
|
4
|
|
|
4
|
|
|
|
|
5582
|
|
87
|
|
|
|
|
|
|
|
88
|
|
|
|
|
|
|
my $offset; # class variable |
89
|
|
|
|
|
|
|
|
90
|
|
|
|
|
|
|
sub new { |
91
|
11
|
|
|
11
|
1
|
40
|
my($class, @args) = @_; |
92
|
11
|
50
|
|
|
|
28
|
$class = ref $class ? ref $class : $class; |
93
|
11
|
|
|
|
|
21
|
my $self = bless {}, $class; |
94
|
11
|
|
|
|
|
35
|
$self->_initialize(@args); |
95
|
11
|
|
|
|
|
49
|
return $self; |
96
|
|
|
|
|
|
|
} |
97
|
|
|
|
|
|
|
|
98
|
|
|
|
|
|
|
|
99
|
|
|
|
|
|
|
{ |
100
|
|
|
|
|
|
|
|
101
|
|
|
|
|
|
|
my %FILE_FORMAT = ( |
102
|
|
|
|
|
|
|
#'itype2' => 'itype2', # itype2 format doesn't work with 'current' |
103
|
|
|
|
|
|
|
#'8' => 'itype2', |
104
|
|
|
|
|
|
|
'withrefm' => 'withrefm', |
105
|
|
|
|
|
|
|
'31' => 'withrefm', |
106
|
|
|
|
|
|
|
#'bairoch' => 'bairoch', # bairoch format doesn't work with 'current' |
107
|
|
|
|
|
|
|
#'19' => 'bairoch', |
108
|
|
|
|
|
|
|
#'macvector' => 'bairoch', |
109
|
|
|
|
|
|
|
#'vectorNTI' => 'bairoch', |
110
|
|
|
|
|
|
|
'neo' => 'neos', |
111
|
|
|
|
|
|
|
'prototype' => 'proto' |
112
|
|
|
|
|
|
|
); |
113
|
|
|
|
|
|
|
|
114
|
|
|
|
|
|
|
sub _initialize { |
115
|
11
|
|
|
11
|
|
21
|
my($self,@args) = @_; |
116
|
11
|
|
|
|
|
78
|
my ($current, $url, $file, $fh, $format, $verbose) = |
117
|
|
|
|
|
|
|
$self->_rearrange([qw(CURRENT URL FILE FH FORMAT VERBOSE)],@args); |
118
|
11
|
50
|
|
|
|
40
|
$verbose || 0; |
119
|
11
|
|
|
|
|
40
|
$self->verbose($verbose); |
120
|
11
|
50
|
33
|
|
|
33
|
if ($current && $format) { |
121
|
0
|
0
|
0
|
|
|
0
|
$self->throw("Can't use -current with file, fh, or url set") if ($url || $file || $fh); |
|
|
|
0
|
|
|
|
|
122
|
0
|
0
|
|
|
|
0
|
$self->throw("Format $format not retrievable using 'current'") if (!exists $FILE_FORMAT{$format}); |
123
|
0
|
|
|
|
|
0
|
my $io = $self->new(-url => 'ftp://ftp.neb.com/pub/rebase/VERSION'); |
124
|
0
|
|
|
|
|
0
|
chomp (my $version = $io->_readline); |
125
|
0
|
|
|
|
|
0
|
push @args, (-url => "ftp://ftp.neb.com/pub/rebase/$FILE_FORMAT{$format}.$version", -retries => 1); |
126
|
|
|
|
|
|
|
} |
127
|
|
|
|
|
|
|
|
128
|
11
|
|
|
|
|
30
|
$self->_companies; |
129
|
11
|
50
|
|
|
|
46
|
return unless $self->SUPER::_initialize(@args); |
130
|
|
|
|
|
|
|
} |
131
|
|
|
|
|
|
|
|
132
|
|
|
|
|
|
|
} |
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
=head2 read |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
Title : read |
137
|
|
|
|
|
|
|
Usage : $renzs = $stream->read |
138
|
|
|
|
|
|
|
Function: reads all the restrction enzymes from the stream |
139
|
|
|
|
|
|
|
Returns : a Bio::Restriction::Restriction object |
140
|
|
|
|
|
|
|
Args : none |
141
|
|
|
|
|
|
|
|
142
|
|
|
|
|
|
|
=cut |
143
|
|
|
|
|
|
|
|
144
|
|
|
|
|
|
|
sub read { |
145
|
5
|
|
|
5
|
1
|
7
|
my $self = shift; |
146
|
|
|
|
|
|
|
|
147
|
5
|
|
|
|
|
27
|
my $renzs = Bio::Restriction::EnzymeCollection->new(-empty => 1); |
148
|
5
|
|
66
|
|
|
52
|
seek DATA,($offset||=tell DATA), 0; |
149
|
5
|
|
|
|
|
63
|
while () { |
150
|
2660
|
|
|
|
|
2488
|
chomp; |
151
|
2660
|
50
|
|
|
|
5667
|
next if /^\s*$/; |
152
|
2660
|
|
|
|
|
7797
|
my ($name, $site, $cut) = split /\s+/; |
153
|
2660
|
|
|
|
|
7536
|
my $re = Bio::Restriction::Enzyme->new(-name => $name, |
154
|
|
|
|
|
|
|
-site => $site, |
155
|
|
|
|
|
|
|
-cut => $cut); |
156
|
2660
|
|
|
|
|
6298
|
$renzs->enzymes($re); |
157
|
|
|
|
|
|
|
} |
158
|
5
|
|
|
|
|
50
|
return $renzs; |
159
|
|
|
|
|
|
|
} |
160
|
|
|
|
|
|
|
|
161
|
|
|
|
|
|
|
=head2 _xln_sub |
162
|
|
|
|
|
|
|
|
163
|
|
|
|
|
|
|
Title : _xln_sub |
164
|
|
|
|
|
|
|
Function: Translates withrefm coords to Bio::Restriction coords |
165
|
|
|
|
|
|
|
Args : Bio::Restriction::Enzyme object, scalar integer (cut posn) |
166
|
|
|
|
|
|
|
Note : Used internally; pass as a coderef to the B:R::Enzyme |
167
|
|
|
|
|
|
|
constructor |
168
|
|
|
|
|
|
|
Note : It is convenient for each format module to have its own |
169
|
|
|
|
|
|
|
version of this; not currently demanded by the interface. |
170
|
|
|
|
|
|
|
=cut |
171
|
|
|
|
|
|
|
|
172
|
|
|
|
|
|
|
sub _xln_sub { # for base.pm, a no-op |
173
|
0
|
|
|
0
|
|
0
|
my ($z,$c) = @_; |
174
|
0
|
|
|
|
|
0
|
return $c; |
175
|
|
|
|
|
|
|
} |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
=head2 write |
179
|
|
|
|
|
|
|
|
180
|
|
|
|
|
|
|
Title : write |
181
|
|
|
|
|
|
|
Usage : $stream->write($renzs) |
182
|
|
|
|
|
|
|
Function: writes restriction enzymes into the stream |
183
|
|
|
|
|
|
|
Returns : 1 for success and 0 for error |
184
|
|
|
|
|
|
|
Args : a Bio::Restriction::Enzyme |
185
|
|
|
|
|
|
|
or a Bio::Restriction::EnzymeCollection object |
186
|
|
|
|
|
|
|
|
187
|
|
|
|
|
|
|
=cut |
188
|
|
|
|
|
|
|
|
189
|
|
|
|
|
|
|
sub write { |
190
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
191
|
0
|
|
|
|
|
0
|
foreach (@_) { |
192
|
0
|
|
|
|
|
0
|
map { printf "%s\t%s\t%s\n", $_->name, $_->string, $_->cut |
193
|
0
|
0
|
|
|
|
0
|
} sort {$a->name cmp $b->name} $_->each_enzyme |
|
0
|
|
|
|
|
0
|
|
194
|
|
|
|
|
|
|
if $_->isa('Bio::Restriction::EnzymeCollection'); |
195
|
0
|
0
|
|
|
|
0
|
printf "%s\t%s\t%s\n", $_->name, $_->string, $_->cut |
196
|
|
|
|
|
|
|
if $_->isa('Bio::Restriction::Enzyme'); |
197
|
|
|
|
|
|
|
} |
198
|
|
|
|
|
|
|
} |
199
|
|
|
|
|
|
|
|
200
|
|
|
|
|
|
|
=head2 verify_prototype |
201
|
|
|
|
|
|
|
|
202
|
|
|
|
|
|
|
Title : verify_prototype |
203
|
|
|
|
|
|
|
Purpose : checks enzyme against current prototype list (retrieved remotely) |
204
|
|
|
|
|
|
|
Returns : returns TRUE if enzyme is prototype |
205
|
|
|
|
|
|
|
Argument : Bio::Restriction::EnzymeI |
206
|
|
|
|
|
|
|
Comments : This is an auxiliary method to retrieve and check an enzyme |
207
|
|
|
|
|
|
|
as a prototype. It retrieves the current list, stores it |
208
|
|
|
|
|
|
|
as a singleton instance, then uses it to check the prototype |
209
|
|
|
|
|
|
|
and modify is_prototype() to true or false. Use as follows: |
210
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
my $col = $io->read; |
212
|
|
|
|
|
|
|
for my $enz ($col->each_enzyme) { |
213
|
|
|
|
|
|
|
print $enz->name.":".$enz->site."\n"; |
214
|
|
|
|
|
|
|
print "\t".$io->verify_prototype($enz)."\n"; |
215
|
|
|
|
|
|
|
} |
216
|
|
|
|
|
|
|
|
217
|
|
|
|
|
|
|
=cut |
218
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
my $protodb; |
220
|
|
|
|
|
|
|
|
221
|
|
|
|
|
|
|
sub verify_prototype { |
222
|
0
|
|
|
0
|
1
|
0
|
my ($self, $enz) = @_; |
223
|
0
|
0
|
0
|
|
|
0
|
$self->throw("Must pass a Bio::Restriction::EnzymeI") unless |
|
|
|
0
|
|
|
|
|
224
|
|
|
|
|
|
|
$enz && ref $enz && $enz->isa("Bio::Restriction::EnzymeI"); |
225
|
0
|
0
|
|
|
|
0
|
if (!defined $protodb) { |
226
|
0
|
|
|
|
|
0
|
my $io = Bio::Restriction::IO->new(-format => 'prototype', |
227
|
|
|
|
|
|
|
-current => 1); |
228
|
0
|
|
|
|
|
0
|
$protodb = $io->read; |
229
|
|
|
|
|
|
|
} |
230
|
0
|
0
|
|
|
|
0
|
if ($protodb->get_enzyme($enz->name)) { |
231
|
0
|
|
|
|
|
0
|
$enz->is_prototype(1); |
232
|
|
|
|
|
|
|
} else { |
233
|
0
|
|
|
|
|
0
|
$enz->is_prototype(0); |
234
|
|
|
|
|
|
|
} |
235
|
0
|
|
|
|
|
0
|
$enz->is_prototype; |
236
|
|
|
|
|
|
|
} |
237
|
|
|
|
|
|
|
|
238
|
|
|
|
|
|
|
=head2 Common REBASE parsing methods |
239
|
|
|
|
|
|
|
|
240
|
|
|
|
|
|
|
The rest of the methods in this file are to be used by other REBASE |
241
|
|
|
|
|
|
|
parsers. They are not to be used outside subclasses of this base |
242
|
|
|
|
|
|
|
class. (They are 'protected' in the sense the word is used in Java.) |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
=cut |
245
|
|
|
|
|
|
|
|
246
|
|
|
|
|
|
|
=head2 _cuts_from_site |
247
|
|
|
|
|
|
|
|
248
|
|
|
|
|
|
|
Title : _cuts_from_site |
249
|
|
|
|
|
|
|
Usage : ($site, $cut, $comp_cut) = _cuts_from_site('ACGCGT(4/5)'); |
250
|
|
|
|
|
|
|
Function: Separates cut positions from a single site string. |
251
|
|
|
|
|
|
|
Does nothing to site if it does not have the cut string |
252
|
|
|
|
|
|
|
Returns : array of site_string, forward_cut_position, reverse_cut_position |
253
|
|
|
|
|
|
|
Args : recognition site string |
254
|
|
|
|
|
|
|
Note : Not used in withrefm refactor/maj |
255
|
|
|
|
|
|
|
|
256
|
|
|
|
|
|
|
=cut |
257
|
|
|
|
|
|
|
|
258
|
|
|
|
|
|
|
sub _cuts_from_site { |
259
|
0
|
|
|
0
|
|
0
|
my ($self, $site) = @_; |
260
|
0
|
|
|
|
|
0
|
my ($cut, $comp_cut) = $site =~ /\((-?\d+)\/(-?\d+)\)/; |
261
|
0
|
|
|
|
|
0
|
$site =~ s/\(.*\)$//; |
262
|
0
|
|
|
|
|
0
|
return ($site, $cut, $comp_cut); |
263
|
|
|
|
|
|
|
} |
264
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
|
266
|
|
|
|
|
|
|
=head2 _meth |
267
|
|
|
|
|
|
|
|
268
|
|
|
|
|
|
|
Title : _meth |
269
|
|
|
|
|
|
|
Usage : ($pos, $meth) = $self->_meth('2(5)'); |
270
|
|
|
|
|
|
|
Function: Separates methylation postion and coce from a string. |
271
|
|
|
|
|
|
|
Adjusts the postion depending on enzyme site length |
272
|
|
|
|
|
|
|
and symmetry |
273
|
|
|
|
|
|
|
Returns : array of position and methylation code |
274
|
|
|
|
|
|
|
Args : 1. reference to Enzyme object |
275
|
|
|
|
|
|
|
2. methylation description string |
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
=cut |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
sub _meth { |
280
|
828
|
|
|
828
|
|
1504
|
my ($self, $re, $meth) = @_; |
281
|
|
|
|
|
|
|
|
282
|
828
|
|
|
|
|
720
|
$meth =~ /(\S+)\((\d+)\)/; |
283
|
828
|
|
|
|
|
1061
|
my ($pos, $m) = ($1, $2); |
284
|
828
|
100
|
|
|
|
1298
|
$pos = 0 if $pos eq '?'; |
285
|
828
|
100
|
100
|
|
|
1790
|
$pos = $re->seq->length + $pos if $pos and ! $re->palindromic; |
286
|
828
|
|
|
|
|
1987
|
return ($pos, $m); |
287
|
|
|
|
|
|
|
|
288
|
0
|
0
|
|
|
|
0
|
$self->warn("Unknown methylation format [$meth]") if $self->verbose >0; |
289
|
|
|
|
|
|
|
} |
290
|
|
|
|
|
|
|
|
291
|
|
|
|
|
|
|
|
292
|
|
|
|
|
|
|
=head2 _coordinate_shift_to_cut |
293
|
|
|
|
|
|
|
|
294
|
|
|
|
|
|
|
Title : _coordinate_shift_to_cut |
295
|
|
|
|
|
|
|
Usage : $cut = $self->_coordinate_shift_to_cut($oricut, offset); |
296
|
|
|
|
|
|
|
Function: Adjust cut position coordinates to start from the |
297
|
|
|
|
|
|
|
first nucleotides of site |
298
|
|
|
|
|
|
|
Returns : Cut position in correct coordinates |
299
|
|
|
|
|
|
|
Args : 1. Original cut position |
300
|
|
|
|
|
|
|
2. Length of the recognition site |
301
|
|
|
|
|
|
|
Note : Not used in withrefm.pm refactor/maj |
302
|
|
|
|
|
|
|
|
303
|
|
|
|
|
|
|
=cut |
304
|
|
|
|
|
|
|
|
305
|
|
|
|
|
|
|
sub _coordinate_shift_to_cut { |
306
|
0
|
|
|
0
|
|
0
|
my ($self, $cut, $site_length) = @_; |
307
|
0
|
|
|
|
|
0
|
return $cut + $site_length; |
308
|
|
|
|
|
|
|
} |
309
|
|
|
|
|
|
|
|
310
|
|
|
|
|
|
|
|
311
|
|
|
|
|
|
|
=head2 _make_multisites |
312
|
|
|
|
|
|
|
|
313
|
|
|
|
|
|
|
Title : _make_multisites |
314
|
|
|
|
|
|
|
Usage : $self->_make_multisites($first_enzyme, \@sites, \@mets) |
315
|
|
|
|
|
|
|
Function: Bless a Bio::Restriction::Enzyme into |
316
|
|
|
|
|
|
|
Bio::Restriction::Enzyme::MultiSite and clone it as many |
317
|
|
|
|
|
|
|
times as there are alternative sites. |
318
|
|
|
|
|
|
|
Returns : nothing, does in place editing |
319
|
|
|
|
|
|
|
Args : 1. a Bio::Restriction::Enzyme |
320
|
|
|
|
|
|
|
2. reference to an array of recognition site strings |
321
|
|
|
|
|
|
|
3. reference to an array of methylation code strings, optional |
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
=cut |
324
|
|
|
|
|
|
|
|
325
|
|
|
|
|
|
|
# removed the enzyme collection from arg list /maj |
326
|
|
|
|
|
|
|
|
327
|
|
|
|
|
|
|
sub _make_multisites { |
328
|
15
|
|
|
15
|
|
23
|
my ($self, $re, $sites, $meths, $xln_sub) = @_; |
329
|
|
|
|
|
|
|
|
330
|
15
|
|
|
|
|
39
|
bless $re, 'Bio::Restriction::Enzyme::MultiSite'; |
331
|
|
|
|
|
|
|
|
332
|
15
|
|
|
|
|
21
|
my $count = 0; |
333
|
15
|
|
|
|
|
21
|
while ($count < scalar @{$sites}) { |
|
41
|
|
|
|
|
96
|
|
334
|
|
|
|
|
|
|
# this should probably be refactored to use the constructor |
335
|
|
|
|
|
|
|
# too, rather than the clone/accessor method /maj |
336
|
|
|
|
|
|
|
# my $re2 = $re->clone; |
337
|
|
|
|
|
|
|
# my $re2; |
338
|
|
|
|
|
|
|
|
339
|
26
|
|
|
|
|
28
|
my $site = @{$sites}[$count]; |
|
26
|
|
|
|
|
44
|
|
340
|
26
|
|
|
|
|
111
|
my ($precut, $recog, $postcut) = ( $site =~ m/^(?:\((\w+\/\w+)\))?([\w^]+)(?:\((\w+\/\w+)\))?/ ); |
341
|
|
|
|
|
|
|
|
342
|
|
|
|
|
|
|
# set the site attribute |
343
|
|
|
|
|
|
|
# $re2->site($recog); |
344
|
|
|
|
|
|
|
|
345
|
|
|
|
|
|
|
# set the recog attribute (which will make the regexp transformation |
346
|
|
|
|
|
|
|
# if necessary: |
347
|
|
|
|
|
|
|
# $re2->recog($recog); |
348
|
|
|
|
|
|
|
# $recog = $re2->string; |
349
|
|
|
|
|
|
|
|
350
|
|
|
|
|
|
|
# no warnings; # avoid 'uninitialized value' warning against $postcut |
351
|
|
|
|
|
|
|
# my ($cut, $comp_cut) = ( $postcut =~ /(-?\d+)\/(-?\d+)/ ); |
352
|
|
|
|
|
|
|
# use warnings; |
353
|
|
|
|
|
|
|
|
354
|
|
|
|
|
|
|
# note the following hard codes the coordinate transformation |
355
|
|
|
|
|
|
|
# used for rebase/itype2 : this method will break on the |
356
|
|
|
|
|
|
|
# base.pm format. |
357
|
|
|
|
|
|
|
# if ($cut) { |
358
|
|
|
|
|
|
|
# $re2->cut($cut + length $recog); |
359
|
|
|
|
|
|
|
# $re2->complementary_cut($comp_cut + length $recog); |
360
|
|
|
|
|
|
|
# } |
361
|
|
|
|
|
|
|
|
362
|
26
|
|
|
|
|
76
|
my $re2 = Bio::Restriction::Enzyme::MultiSite->new( |
363
|
|
|
|
|
|
|
-name => $re->name, |
364
|
|
|
|
|
|
|
-site => $recog, |
365
|
|
|
|
|
|
|
-recog => $recog, |
366
|
|
|
|
|
|
|
-precut => $precut, |
367
|
|
|
|
|
|
|
-postcut => $postcut, |
368
|
|
|
|
|
|
|
-xln_sub => $xln_sub |
369
|
|
|
|
|
|
|
); |
370
|
|
|
|
|
|
|
|
371
|
26
|
100
|
66
|
|
|
106
|
if ($meths and @$meths) { |
372
|
22
|
|
|
|
|
48
|
$re2->purge_methylation_sites; |
373
|
22
|
|
|
|
|
36
|
$re2->methylation_sites($self->_meth($re2, @{$meths}[$count])); |
|
22
|
|
|
|
|
43
|
|
374
|
|
|
|
|
|
|
} |
375
|
|
|
|
|
|
|
|
376
|
26
|
|
|
|
|
59
|
$re->others($re2); |
377
|
26
|
|
|
|
|
26
|
$count++; |
378
|
|
|
|
|
|
|
} |
379
|
|
|
|
|
|
|
|
380
|
15
|
|
|
|
|
58
|
foreach my $enz ($re->others) { |
381
|
26
|
|
|
|
|
41
|
$enz->others($re, grep {$_ ne $enz} $re->others); |
|
66
|
|
|
|
|
114
|
|
382
|
|
|
|
|
|
|
} |
383
|
|
|
|
|
|
|
|
384
|
15
|
|
|
|
|
51
|
1; |
385
|
|
|
|
|
|
|
} |
386
|
|
|
|
|
|
|
|
387
|
|
|
|
|
|
|
=head2 _make_multicuts |
388
|
|
|
|
|
|
|
|
389
|
|
|
|
|
|
|
Title : _make_multicuts |
390
|
|
|
|
|
|
|
Usage : $self->_make_multicuts($first_enzyme, $precuts) |
391
|
|
|
|
|
|
|
Function: |
392
|
|
|
|
|
|
|
|
393
|
|
|
|
|
|
|
Bless a Bio::Restriction::Enzyme into |
394
|
|
|
|
|
|
|
Bio::Restriction::Enzyme::MultiCut and clone it. The precut |
395
|
|
|
|
|
|
|
string is processed to replase the cut sites in the cloned |
396
|
|
|
|
|
|
|
object. Both objects refer to each other through others() method. |
397
|
|
|
|
|
|
|
|
398
|
|
|
|
|
|
|
Returns : nothing, does in place editing |
399
|
|
|
|
|
|
|
Args : 1. a Bio::Restriction::Enzyme |
400
|
|
|
|
|
|
|
2. precut string, e.g. '12/7' |
401
|
|
|
|
|
|
|
|
402
|
|
|
|
|
|
|
|
403
|
|
|
|
|
|
|
The examples we have of multiply cutting enzymes cut only four |
404
|
|
|
|
|
|
|
times. This protected method deals only with a string of two |
405
|
|
|
|
|
|
|
integers separated with a slash, e.g. '12/7'. The numbers represent the postions |
406
|
|
|
|
|
|
|
BEFORE the start of the recognition site, i.e. negative positions. |
407
|
|
|
|
|
|
|
|
408
|
|
|
|
|
|
|
=cut |
409
|
|
|
|
|
|
|
|
410
|
|
|
|
|
|
|
# removed the enzyme collection from arg list /maj |
411
|
|
|
|
|
|
|
|
412
|
|
|
|
|
|
|
sub _make_multicuts { |
413
|
0
|
|
|
0
|
|
0
|
my ($self, $re, $precut) = @_; |
414
|
|
|
|
|
|
|
|
415
|
0
|
|
|
|
|
0
|
bless $re, 'Bio::Restriction::Enzyme::MultiCut'; |
416
|
|
|
|
|
|
|
|
417
|
0
|
|
|
|
|
0
|
my ($cut, $comp_cut) = $precut =~ /(-?\d+)\/(-?\d+)/; |
418
|
|
|
|
|
|
|
|
419
|
0
|
|
|
|
|
0
|
my $re2 = $re->clone; |
420
|
|
|
|
|
|
|
|
421
|
0
|
|
|
|
|
0
|
$re2->cut("-$cut"); |
422
|
0
|
|
|
|
|
0
|
$re2->complementary_cut("-$comp_cut"); |
423
|
|
|
|
|
|
|
|
424
|
0
|
|
|
|
|
0
|
$re->others($re2); |
425
|
|
|
|
|
|
|
|
426
|
0
|
|
|
|
|
0
|
1; |
427
|
|
|
|
|
|
|
} |
428
|
|
|
|
|
|
|
|
429
|
|
|
|
|
|
|
=head2 _companies |
430
|
|
|
|
|
|
|
|
431
|
|
|
|
|
|
|
Title : _companies |
432
|
|
|
|
|
|
|
Purpose : Defines the companies that we know about |
433
|
|
|
|
|
|
|
Returns : A hash |
434
|
|
|
|
|
|
|
Argument : Nothing |
435
|
|
|
|
|
|
|
Comments : An internal method to define the companies that we know about |
436
|
|
|
|
|
|
|
REBASE uses a code, and this converts the code to the real name |
437
|
|
|
|
|
|
|
(e.g. A = Amersham Pharmacia Biotech) |
438
|
|
|
|
|
|
|
|
439
|
|
|
|
|
|
|
=cut |
440
|
|
|
|
|
|
|
|
441
|
|
|
|
|
|
|
sub _companies { |
442
|
|
|
|
|
|
|
# this is just so it is easy to set up the codes that REBASE uses |
443
|
11
|
|
|
11
|
|
15
|
my $self=shift; |
444
|
11
|
|
|
|
|
137
|
my %companies=( |
445
|
|
|
|
|
|
|
'A'=>'Amersham Pharmacia Biotech (1/03)', |
446
|
|
|
|
|
|
|
'C'=>'Minotech Biotechnology (6/01)', |
447
|
|
|
|
|
|
|
'E'=>'Stratagene (1/03)', |
448
|
|
|
|
|
|
|
'F'=>'Fermentas AB (1/03)', |
449
|
|
|
|
|
|
|
'G'=>'Qbiogene (1/03)', |
450
|
|
|
|
|
|
|
'H'=>'American Allied Biochemical, Inc. (10/98)', |
451
|
|
|
|
|
|
|
'I'=>'SibEnzyme Ltd. (1/03)', |
452
|
|
|
|
|
|
|
'J'=>'Nippon Gene Co., Ltd. (6/00)', |
453
|
|
|
|
|
|
|
'K'=>'Takara Shuzo Co. Ltd. (1/03)', |
454
|
|
|
|
|
|
|
'M'=>'Roche Applied Science (1/03)', |
455
|
|
|
|
|
|
|
'N'=>'New England Biolabs (1/03)', |
456
|
|
|
|
|
|
|
'O'=>'Toyobo Biochemicals (11/98)', |
457
|
|
|
|
|
|
|
'P'=>'Megabase Research Products (5/99)', |
458
|
|
|
|
|
|
|
'Q'=>'CHIMERx (1/03)', |
459
|
|
|
|
|
|
|
'R'=>'Promega Corporation (1/03)', |
460
|
|
|
|
|
|
|
'S'=>'Sigma Chemical Corporation (1/03)', |
461
|
|
|
|
|
|
|
'U'=>'Bangalore Genei (1/03)', |
462
|
|
|
|
|
|
|
'V'=>'MRC-Holland (1/03)', |
463
|
|
|
|
|
|
|
'X'=>'EURx Ltd. (1/03)'); |
464
|
11
|
|
|
|
|
22
|
$self->{company}=\%companies; |
465
|
|
|
|
|
|
|
} |
466
|
|
|
|
|
|
|
|
467
|
|
|
|
|
|
|
1; |
468
|
|
|
|
|
|
|
|
469
|
|
|
|
|
|
|
__DATA__ |