line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
# $Id: Term.pm 2013-06-06 erick.antezana $ |
2
|
|
|
|
|
|
|
# |
3
|
|
|
|
|
|
|
# Module : Term.pm |
4
|
|
|
|
|
|
|
# Purpose : Term of an Ontology. |
5
|
|
|
|
|
|
|
# License : Copyright (c) 2006-2015 by Erick Antezana. All rights reserved. |
6
|
|
|
|
|
|
|
# This program is free software; you can redistribute it and/or |
7
|
|
|
|
|
|
|
# modify it under the same terms as Perl itself. |
8
|
|
|
|
|
|
|
# Contact : Erick Antezana |
9
|
|
|
|
|
|
|
# |
10
|
|
|
|
|
|
|
package OBO::Core::Term; |
11
|
|
|
|
|
|
|
|
12
|
12
|
|
|
12
|
|
14314
|
use OBO::Util::InstanceSet; |
|
12
|
|
|
|
|
24
|
|
|
12
|
|
|
|
|
372
|
|
13
|
12
|
|
|
12
|
|
3779
|
use OBO::Core::Synonym; |
|
12
|
|
|
|
|
26
|
|
|
12
|
|
|
|
|
316
|
|
14
|
12
|
|
|
12
|
|
3937
|
use OBO::Util::SynonymSet; |
|
12
|
|
|
|
|
27
|
|
|
12
|
|
|
|
|
275
|
|
15
|
|
|
|
|
|
|
|
16
|
12
|
|
|
12
|
|
60
|
use Carp; |
|
12
|
|
|
|
|
24
|
|
|
12
|
|
|
|
|
668
|
|
17
|
12
|
|
|
12
|
|
59
|
use strict; |
|
12
|
|
|
|
|
24
|
|
|
12
|
|
|
|
|
227
|
|
18
|
12
|
|
|
12
|
|
56
|
use warnings; |
|
12
|
|
|
|
|
21
|
|
|
12
|
|
|
|
|
47566
|
|
19
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
sub new { |
21
|
2180
|
|
|
2180
|
0
|
3358
|
my $class = shift; |
22
|
2180
|
|
|
|
|
3277
|
my $self = {}; |
23
|
|
|
|
|
|
|
|
24
|
2180
|
|
|
|
|
4684
|
$self->{ID} = undef; # required, scalar (1) |
25
|
2180
|
|
|
|
|
3340
|
$self->{IS_ANONYMOUS} = undef; # [1|0], 0 by default |
26
|
2180
|
|
|
|
|
3542
|
$self->{NAME} = undef; # not required since OBO spec 1.4, scalar (0..1) |
27
|
2180
|
|
|
|
|
6177
|
$self->{NAMESPACE_SET} = OBO::Util::Set->new(); # set (0..N) |
28
|
2180
|
|
|
|
|
5502
|
$self->{ALT_ID} = OBO::Util::Set->new(); # set (0..N) |
29
|
2180
|
|
|
|
|
3517
|
$self->{BUILTIN} = undef; # [1|0], 0 by default |
30
|
2180
|
|
|
|
|
6081
|
$self->{DEF} = OBO::Core::Def->new(); # (0..1) |
31
|
2180
|
|
|
|
|
5169
|
$self->{COMMENT} = undef; # scalar (0..1) |
32
|
2180
|
|
|
|
|
5560
|
$self->{SUBSET_SET} = OBO::Util::Set->new(); # set of scalars (0..N) |
33
|
2180
|
|
|
|
|
5993
|
$self->{SYNONYM_SET} = OBO::Util::SynonymSet->new(); # set of synonyms (0..N) |
34
|
2180
|
|
|
|
|
5930
|
$self->{XREF_SET} = OBO::Util::DbxrefSet->new(); # set of dbxref's (0..N) |
35
|
2180
|
|
|
|
|
5709
|
$self->{PROPERTY_VALUE} = OBO::Util::ObjectSet->new(); # set of objects: rel's Term->Instance or Term->Datatype (0..N) |
36
|
2180
|
|
|
|
|
5945
|
$self->{CLASS_OF} = OBO::Util::InstanceSet->new();# set of instances (0..N) |
37
|
2180
|
|
|
|
|
5865
|
$self->{INTERSECTION_OF} = OBO::Util::Set->new(); # (0..N) with N=0, 2, 3, ... |
38
|
2180
|
|
|
|
|
5546
|
$self->{UNION_OF} = OBO::Util::Set->new(); # (0..N) with N=0, 2, 3, ... |
39
|
2180
|
|
|
|
|
5401
|
$self->{DISJOINT_FROM} = OBO::Util::Set->new(); # (0..N) |
40
|
2180
|
|
|
|
|
3438
|
$self->{CREATED_BY} = undef; # scalar (0..1) |
41
|
2180
|
|
|
|
|
3683
|
$self->{CREATION_DATE} = undef; # scalar (0..1) |
42
|
2180
|
|
|
|
|
3467
|
$self->{MODIFIED_BY} = undef; # scalar (0..1) |
43
|
2180
|
|
|
|
|
3063
|
$self->{MODIFICATION_DATE} = undef; # scalar (0..1) |
44
|
2180
|
|
|
|
|
2930
|
$self->{IS_OBSOLETE} = undef; # [1|0], 0 by default |
45
|
2180
|
|
|
|
|
5559
|
$self->{REPLACED_BY} = OBO::Util::Set->new(); # set of scalars (0..N) |
46
|
2180
|
|
|
|
|
6070
|
$self->{CONSIDER} = OBO::Util::Set->new(); # set of scalars (0..N) |
47
|
|
|
|
|
|
|
|
48
|
2180
|
|
|
|
|
3244
|
bless ($self, $class); |
49
|
2180
|
|
|
|
|
4632
|
return $self; |
50
|
|
|
|
|
|
|
} |
51
|
|
|
|
|
|
|
|
52
|
|
|
|
|
|
|
=head2 id |
53
|
|
|
|
|
|
|
|
54
|
|
|
|
|
|
|
Usage - print $term->id() or $term->id($id) |
55
|
|
|
|
|
|
|
Returns - the term ID (string) |
56
|
|
|
|
|
|
|
Args - the term ID (string) |
57
|
|
|
|
|
|
|
Function - gets/sets the ID of this term |
58
|
|
|
|
|
|
|
|
59
|
|
|
|
|
|
|
=cut |
60
|
|
|
|
|
|
|
|
61
|
|
|
|
|
|
|
sub id { |
62
|
227259
|
100
|
|
227259
|
1
|
445936
|
if ($_[1]) { $_[0]->{ID} = $_[1] } |
|
2182
|
|
|
|
|
4554
|
|
63
|
227259
|
|
|
|
|
657970
|
return $_[0]->{ID}; |
64
|
|
|
|
|
|
|
} |
65
|
|
|
|
|
|
|
|
66
|
|
|
|
|
|
|
=head2 idspace |
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
Usage - print $term->idspace() |
69
|
|
|
|
|
|
|
Returns - the idspace of this term; otherwise, 'NN' |
70
|
|
|
|
|
|
|
Args - none |
71
|
|
|
|
|
|
|
Function - gets the idspace of this term # TODO Does this method still makes sense? |
72
|
|
|
|
|
|
|
|
73
|
|
|
|
|
|
|
=cut |
74
|
|
|
|
|
|
|
|
75
|
|
|
|
|
|
|
sub idspace { |
76
|
10
|
100
|
|
10
|
1
|
121
|
$_[0]->{ID} =~ /([A-Za-z_]+):/ if ($_[0]->{ID}); |
77
|
10
|
|
100
|
|
|
94
|
return $1 || 'NN'; |
78
|
|
|
|
|
|
|
} |
79
|
|
|
|
|
|
|
|
80
|
|
|
|
|
|
|
=head2 subnamespace |
81
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
Usage - print $term->subnamespace() |
83
|
|
|
|
|
|
|
Returns - the subnamespace of this term (character); otherwise, 'X' |
84
|
|
|
|
|
|
|
Args - none |
85
|
|
|
|
|
|
|
Function - gets the subnamespace of this term |
86
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
=cut |
88
|
|
|
|
|
|
|
|
89
|
|
|
|
|
|
|
sub subnamespace { |
90
|
2
|
100
|
|
2
|
1
|
11
|
$_[0]->{ID} =~ /:([A-Z][a-z]?)/ if ($_[0]->{ID}); |
91
|
2
|
|
100
|
|
|
15
|
return $1 || 'X'; |
92
|
|
|
|
|
|
|
} |
93
|
|
|
|
|
|
|
|
94
|
|
|
|
|
|
|
=head2 code |
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
Usage - print $term->code() |
97
|
|
|
|
|
|
|
Returns - the code of this term (character); otherwise, '0000000' |
98
|
|
|
|
|
|
|
Args - none |
99
|
|
|
|
|
|
|
Function - gets the code of this term |
100
|
|
|
|
|
|
|
|
101
|
|
|
|
|
|
|
=cut |
102
|
|
|
|
|
|
|
|
103
|
|
|
|
|
|
|
sub code { |
104
|
2
|
100
|
|
2
|
1
|
9
|
$_[0]->{ID} =~ /:[A-Z]?[a-z]?(.*)/ if ($_[0]->{ID}); |
105
|
2
|
|
100
|
|
|
13
|
return $1 || '0000000'; |
106
|
|
|
|
|
|
|
} |
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
=head2 name |
109
|
|
|
|
|
|
|
|
110
|
|
|
|
|
|
|
Usage - print $term->name() or $term->name($name) |
111
|
|
|
|
|
|
|
Returns - the name (string) of this term |
112
|
|
|
|
|
|
|
Args - the name (string) of this term |
113
|
|
|
|
|
|
|
Function - gets/sets the name of this term |
114
|
|
|
|
|
|
|
|
115
|
|
|
|
|
|
|
=cut |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
sub name { |
118
|
15650
|
100
|
|
15650
|
1
|
31538
|
if ($_[1]) { $_[0]->{NAME} = $_[1] } |
|
2165
|
|
|
|
|
4680
|
|
119
|
15650
|
|
|
|
|
50777
|
return $_[0]->{NAME}; |
120
|
|
|
|
|
|
|
} |
121
|
|
|
|
|
|
|
|
122
|
|
|
|
|
|
|
=head2 is_anonymous |
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
Usage - print $term->is_anonymous() or $term->is_anonymous("1") |
125
|
|
|
|
|
|
|
Returns - either 1 (true) or 0 (false) |
126
|
|
|
|
|
|
|
Args - either 1 (true) or 0 (false) |
127
|
|
|
|
|
|
|
Function - tells whether this term is anonymous or not. |
128
|
|
|
|
|
|
|
|
129
|
|
|
|
|
|
|
=cut |
130
|
|
|
|
|
|
|
|
131
|
|
|
|
|
|
|
sub is_anonymous { |
132
|
2756
|
50
|
66
|
2756
|
1
|
6492
|
if (defined $_[1] && ($_[1] == 1 || $_[1] == 0)) { $_[0]->{IS_ANONYMOUS} = $_[1] } |
|
12
|
|
66
|
|
|
25
|
|
133
|
2756
|
100
|
66
|
|
|
13618
|
return ($_[0]->{IS_ANONYMOUS} && $_[0]->{IS_ANONYMOUS} == 1)?1:0; |
134
|
|
|
|
|
|
|
} |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
=head2 alt_id |
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
Usage - $term->alt_id() or $term->alt_id($id1, $id2, $id3, ...) |
139
|
|
|
|
|
|
|
Returns - a set (OBO::Util::Set) with the alternate id(s) of this term |
140
|
|
|
|
|
|
|
Args - the alternate id(s) (string) of this term |
141
|
|
|
|
|
|
|
Function - gets/sets the alternate id(s) of this term |
142
|
|
|
|
|
|
|
|
143
|
|
|
|
|
|
|
=cut |
144
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
sub alt_id { |
146
|
3404
|
|
|
3404
|
1
|
4647
|
my $self = shift; |
147
|
3404
|
100
|
|
|
|
9173
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
148
|
1
|
|
|
|
|
6
|
$self->{ALT_ID}->add_all(@_); |
149
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
150
|
11
|
|
|
|
|
46
|
$self->{ALT_ID}->add(shift); |
151
|
|
|
|
|
|
|
} |
152
|
3404
|
|
|
|
|
11458
|
return $self->{ALT_ID}; |
153
|
|
|
|
|
|
|
} |
154
|
|
|
|
|
|
|
|
155
|
|
|
|
|
|
|
=head2 def |
156
|
|
|
|
|
|
|
|
157
|
|
|
|
|
|
|
Usage - $term->def() or $term->def($def) |
158
|
|
|
|
|
|
|
Returns - the definition (OBO::Core::Def) of this term |
159
|
|
|
|
|
|
|
Args - the definition (OBO::Core::Def) of this term |
160
|
|
|
|
|
|
|
Function - gets/sets the definition of the term |
161
|
|
|
|
|
|
|
|
162
|
|
|
|
|
|
|
=cut |
163
|
|
|
|
|
|
|
|
164
|
|
|
|
|
|
|
sub def { |
165
|
7747
|
100
|
|
7747
|
1
|
16208
|
$_[0]->{DEF} = $_[1] if ($_[1]); |
166
|
7747
|
|
|
|
|
25979
|
return $_[0]->{DEF}; |
167
|
|
|
|
|
|
|
} |
168
|
|
|
|
|
|
|
|
169
|
|
|
|
|
|
|
=head2 def_as_string |
170
|
|
|
|
|
|
|
|
171
|
|
|
|
|
|
|
Usage - $term->def_as_string() or $term->def_as_string("During meiosis, the synthesis of DNA proceeding from the broken 3' single-strand DNA end that uses the homologous intact duplex as the template.", "[GOC:elh, PMID:9334324]") |
172
|
|
|
|
|
|
|
Returns - the definition (string) of this term |
173
|
|
|
|
|
|
|
Args - the definition (string) of this term plus the dbxref list (string) describing the source of this definition |
174
|
|
|
|
|
|
|
Function - gets/sets the definition of this term |
175
|
|
|
|
|
|
|
Remark - make sure that colons (,) are scaped (\,) when necessary |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
=cut |
178
|
|
|
|
|
|
|
|
179
|
|
|
|
|
|
|
sub def_as_string { |
180
|
544
|
|
|
544
|
1
|
700
|
my $dbxref_as_string = $_[2]; |
181
|
544
|
100
|
66
|
|
|
1345
|
if (defined $_[1] && defined $dbxref_as_string) { |
182
|
8
|
|
|
|
|
17
|
my $def = $_[0]->{DEF}; |
183
|
8
|
|
|
|
|
33
|
$def->text($_[1]); |
184
|
8
|
|
|
|
|
33
|
my $dbxref_set = OBO::Util::DbxrefSet->new(); |
185
|
|
|
|
|
|
|
|
186
|
8
|
|
|
|
|
30
|
my ($e, $entry) = __dbxref($dbxref_set, $dbxref_as_string); |
187
|
8
|
50
|
|
|
|
26
|
if ($e == -1) { |
188
|
0
|
|
|
|
|
0
|
croak "ERROR: Check the 'dbxref' field of '", $entry, "' (term ID = ", $_[0]->id(), ")." ; |
189
|
|
|
|
|
|
|
} |
190
|
|
|
|
|
|
|
|
191
|
8
|
|
|
|
|
38
|
$def->dbxref_set($dbxref_set); |
192
|
|
|
|
|
|
|
} |
193
|
|
|
|
|
|
|
|
194
|
470
|
|
|
|
|
1001
|
my @sorted_dbxrefs = map { $_->[0] } # restore original values |
195
|
113
|
|
|
|
|
230
|
sort { $a->[1] cmp $b->[1] } # sort |
196
|
470
|
|
|
|
|
1192
|
map { [$_, lc($_->id())] } # transform: value, sortkey |
197
|
544
|
|
|
|
|
1523
|
$_[0]->{DEF}->dbxref_set()->get_set(); |
198
|
|
|
|
|
|
|
|
199
|
544
|
|
|
|
|
1124
|
my @result = (); # a Set? |
200
|
544
|
|
|
|
|
852
|
foreach my $dbxref (@sorted_dbxrefs) { |
201
|
470
|
|
|
|
|
1298
|
push @result, $dbxref->as_string(); |
202
|
|
|
|
|
|
|
} |
203
|
544
|
|
|
|
|
1649
|
my $d = $_[0]->{DEF}->text(); |
204
|
544
|
100
|
|
|
|
1160
|
if (defined $d) { |
205
|
539
|
|
|
|
|
1465
|
return '"'.$_[0]->{DEF}->text().'"'.' ['.join(', ', @result).']'; |
206
|
|
|
|
|
|
|
} else { |
207
|
5
|
|
|
|
|
31
|
return '"" ['.join(', ', @result).']'; |
208
|
|
|
|
|
|
|
} |
209
|
|
|
|
|
|
|
} |
210
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
=head2 namespace |
212
|
|
|
|
|
|
|
|
213
|
|
|
|
|
|
|
Usage - $term->namespace() or $term->namespace($ns1, $ns2, $ns3, ...) |
214
|
|
|
|
|
|
|
Returns - an array with the namespace(s) to which this term belongs |
215
|
|
|
|
|
|
|
Args - the namespace(s) to which this term belongs |
216
|
|
|
|
|
|
|
Function - gets/sets the namespace(s) to which this term belongs |
217
|
|
|
|
|
|
|
|
218
|
|
|
|
|
|
|
=cut |
219
|
|
|
|
|
|
|
|
220
|
|
|
|
|
|
|
sub namespace { |
221
|
2116
|
|
|
2116
|
1
|
2919
|
my $self = shift; |
222
|
2116
|
50
|
|
|
|
6113
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
223
|
0
|
|
|
|
|
0
|
$self->{NAMESPACE_SET}->add_all(@_); |
224
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
225
|
15
|
|
|
|
|
50
|
$self->{NAMESPACE_SET}->add(shift); |
226
|
|
|
|
|
|
|
} |
227
|
2116
|
|
|
|
|
7181
|
return $self->{NAMESPACE_SET}->get_set(); |
228
|
|
|
|
|
|
|
} |
229
|
|
|
|
|
|
|
|
230
|
|
|
|
|
|
|
=head2 comment |
231
|
|
|
|
|
|
|
|
232
|
|
|
|
|
|
|
Usage - print $term->comment() or $term->comment("This is a comment") |
233
|
|
|
|
|
|
|
Returns - the comment (string) of this term |
234
|
|
|
|
|
|
|
Args - the comment (string) of this term |
235
|
|
|
|
|
|
|
Function - gets/sets the comment of this term |
236
|
|
|
|
|
|
|
|
237
|
|
|
|
|
|
|
=cut |
238
|
|
|
|
|
|
|
|
239
|
|
|
|
|
|
|
sub comment { |
240
|
3982
|
100
|
|
3982
|
1
|
8101
|
if (defined $_[1]) { $_[0]->{COMMENT} = $_[1] } |
|
194
|
|
|
|
|
448
|
|
241
|
3982
|
|
|
|
|
12452
|
return $_[0]->{COMMENT}; |
242
|
|
|
|
|
|
|
} |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
=head2 subset |
245
|
|
|
|
|
|
|
|
246
|
|
|
|
|
|
|
Usage - $term->subset() or $term->subset($ss_name1, $ss_name2, $ss_name3, ...) |
247
|
|
|
|
|
|
|
Returns - an array with the subset name(s) to which this term belongs |
248
|
|
|
|
|
|
|
Args - the subset name(s) (string) to which this term belongs |
249
|
|
|
|
|
|
|
Function - gets/sets the subset name(s) to which this term belongs |
250
|
|
|
|
|
|
|
|
251
|
|
|
|
|
|
|
=cut |
252
|
|
|
|
|
|
|
|
253
|
|
|
|
|
|
|
sub subset { |
254
|
3436
|
|
|
3436
|
1
|
4591
|
my $self = shift; |
255
|
3436
|
100
|
|
|
|
9454
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
256
|
1
|
|
|
|
|
4
|
$self->{SUBSET_SET}->add_all(@_); |
257
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
258
|
37
|
|
|
|
|
124
|
$self->{SUBSET_SET}->add(shift); |
259
|
|
|
|
|
|
|
} |
260
|
3436
|
|
|
|
|
10878
|
return $self->{SUBSET_SET}->get_set(); |
261
|
|
|
|
|
|
|
} |
262
|
|
|
|
|
|
|
|
263
|
|
|
|
|
|
|
=head2 synonym_set |
264
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
Usage - $term->synonym_set() or $term->synonym_set($synonym1, $synonym2, $synonym3, ...) |
266
|
|
|
|
|
|
|
Returns - an array with the synonym(s) of this term |
267
|
|
|
|
|
|
|
Args - the synonym(s) (OBO::Core::Synonym) of this term |
268
|
|
|
|
|
|
|
Function - gets/sets the synonym(s) of this term |
269
|
|
|
|
|
|
|
Remark1 - if the synonym (text) is already in the set of synonyms of this term, its scope (and their dbxref's) will be updated (provided they have the same synonym type name) |
270
|
|
|
|
|
|
|
Remark2 - a synonym text identical to the term name is not added to the set of synonyms of this term |
271
|
|
|
|
|
|
|
|
272
|
|
|
|
|
|
|
=cut |
273
|
|
|
|
|
|
|
|
274
|
|
|
|
|
|
|
sub synonym_set { |
275
|
5550
|
|
|
5550
|
1
|
7288
|
my $self = shift; |
276
|
5550
|
|
|
|
|
8888
|
foreach my $synonym (@_) { |
277
|
2128
|
|
|
|
|
4270
|
my $term_name = $self->name(); |
278
|
2128
|
50
|
|
|
|
4948
|
if (!defined($term_name)) { |
279
|
0
|
|
|
|
|
0
|
croak 'The name of this term (', $self->id(), ') is undefined. Add it before adding its synonyms.'; |
280
|
|
|
|
|
|
|
} |
281
|
|
|
|
|
|
|
|
282
|
|
|
|
|
|
|
# |
283
|
|
|
|
|
|
|
# update the scope (and dbxref's) of a synonym -- if the text and synonym type name are identical in both synonyms |
284
|
|
|
|
|
|
|
# |
285
|
2128
|
|
|
|
|
2440
|
my $syn_found = 0; |
286
|
2128
|
|
|
|
|
6017
|
foreach my $s ($self->{SYNONYM_SET}->get_set()) { |
287
|
|
|
|
|
|
|
|
288
|
3715
|
100
|
|
|
|
9009
|
if ($s->def()->text() eq $synonym->def()->text()) { # if that SYNONYM is already in the set |
289
|
|
|
|
|
|
|
|
290
|
9
|
|
|
|
|
26
|
my $synonym_type_name = $synonym->synonym_type_name(); |
291
|
9
|
|
|
|
|
25
|
my $s_type_name = $s->synonym_type_name(); |
292
|
9
|
100
|
66
|
|
|
34
|
if ($synonym_type_name || $s_type_name) { # if any of their STN's is defined |
293
|
6
|
100
|
66
|
|
|
50
|
if ($s_type_name && $synonym_type_name && ($s_type_name eq $synonym_type_name)) { # they should be identical |
|
|
|
100
|
|
|
|
|
294
|
|
|
|
|
|
|
|
295
|
1
|
|
|
|
|
5
|
$s->def()->dbxref_set($synonym->def()->dbxref_set); # then update its DBXREFs! |
296
|
1
|
|
|
|
|
4
|
$s->scope($synonym->scope); # then update its SCOPE! |
297
|
|
|
|
|
|
|
|
298
|
1
|
|
|
|
|
2
|
$syn_found = 1; |
299
|
1
|
|
|
|
|
2
|
last; |
300
|
|
|
|
|
|
|
} |
301
|
|
|
|
|
|
|
} else { |
302
|
3
|
|
|
|
|
10
|
$s->def()->dbxref_set($synonym->def()->dbxref_set); # then update its DBXREFs! |
303
|
3
|
|
|
|
|
10
|
$s->scope($synonym->scope); # then update its SCOPE! |
304
|
|
|
|
|
|
|
|
305
|
3
|
|
|
|
|
4
|
$syn_found = 1; |
306
|
3
|
|
|
|
|
6
|
last; |
307
|
|
|
|
|
|
|
} |
308
|
|
|
|
|
|
|
} |
309
|
|
|
|
|
|
|
} |
310
|
|
|
|
|
|
|
|
311
|
|
|
|
|
|
|
# do not add 'EXACT' synonyms with the same 'name': |
312
|
2128
|
50
|
66
|
|
|
8887
|
if (!$syn_found && !($synonym->scope() eq 'EXACT' && $synonym->def()->text() eq $term_name)) { |
|
|
|
66
|
|
|
|
|
313
|
2124
|
50
|
|
|
|
5644
|
$self->{SYNONYM_SET}->add($synonym) || warn "ERROR: the synonym (", $synonym->def()->text(), ") was not added!!"; |
314
|
|
|
|
|
|
|
} |
315
|
|
|
|
|
|
|
} |
316
|
5550
|
|
|
|
|
16759
|
return $self->{SYNONYM_SET}->get_set(); |
317
|
|
|
|
|
|
|
} |
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
=head2 synonym_as_string |
320
|
|
|
|
|
|
|
|
321
|
|
|
|
|
|
|
Usage - print $term->synonym_as_string() or $term->synonym_as_string('this is a synonym text', '[APO:ea]', 'EXACT', 'UK_SPELLING') |
322
|
|
|
|
|
|
|
Returns - an array with the synonym(s) of this term |
323
|
|
|
|
|
|
|
Args - the synonym text (string), the dbxrefs (string), synonym scope (string) of this term, and optionally the synonym type name (string) |
324
|
|
|
|
|
|
|
Function - gets/sets the synonym(s) of this term |
325
|
|
|
|
|
|
|
Remark1 - if the synonym (text) is already in the set of synonyms of this term, its scope (and their dbxref's) will be updated (provided they have the same synonym type name) |
326
|
|
|
|
|
|
|
Remark2 - a synonym text identical to the term name is not added to the set of synonyms of this term |
327
|
|
|
|
|
|
|
|
328
|
|
|
|
|
|
|
=cut |
329
|
|
|
|
|
|
|
|
330
|
|
|
|
|
|
|
sub synonym_as_string { |
331
|
2141
|
50
|
66
|
2141
|
1
|
12531
|
if ($_[1] && $_[2] && $_[3]) { |
|
|
|
33
|
|
|
|
|
332
|
2125
|
|
|
|
|
6336
|
my $synonym = OBO::Core::Synonym->new(); |
333
|
2125
|
|
|
|
|
5952
|
$synonym->def_as_string($_[1], $_[2]); |
334
|
2125
|
|
|
|
|
7425
|
$synonym->scope($_[3]); |
335
|
2125
|
|
|
|
|
5853
|
$synonym->synonym_type_name($_[4]); # optional argument |
336
|
2125
|
|
|
|
|
4950
|
$_[0]->synonym_set($synonym); |
337
|
|
|
|
|
|
|
} |
338
|
|
|
|
|
|
|
|
339
|
5880
|
|
|
|
|
11392
|
my @sorted_syns = map { $_->[0] } # restore original values |
340
|
6158
|
|
|
|
|
9499
|
sort { $a->[1] cmp $b->[1] } # sort |
341
|
5880
|
|
|
|
|
15294
|
map { [$_, lc($_->def_as_string())] } # transform: value, sortkey |
342
|
2141
|
|
|
|
|
7185
|
$_[0]->{SYNONYM_SET}->get_set(); |
343
|
|
|
|
|
|
|
|
344
|
2141
|
|
|
|
|
5891
|
my @result; |
345
|
|
|
|
|
|
|
my $s_as_string; |
346
|
2141
|
|
|
|
|
3449
|
foreach my $synonym (@sorted_syns) { |
347
|
5880
|
|
|
|
|
14765
|
my $syn_scope = $synonym->scope(); |
348
|
5880
|
50
|
|
|
|
11370
|
if ($syn_scope) { |
349
|
5880
|
|
|
|
|
13929
|
my $syn_type_name = $synonym->synonym_type_name(); |
350
|
5880
|
100
|
|
|
|
9662
|
if ($syn_type_name) { |
351
|
26
|
|
|
|
|
66
|
$s_as_string = ' '.$syn_scope.' '.$syn_type_name; |
352
|
|
|
|
|
|
|
} else { |
353
|
5854
|
|
|
|
|
9360
|
$s_as_string = ' '.$syn_scope; |
354
|
|
|
|
|
|
|
} |
355
|
|
|
|
|
|
|
} else { |
356
|
|
|
|
|
|
|
# This case should never happen since the SCOPE is mandatory! |
357
|
0
|
|
|
|
|
0
|
warn "The scope of this synonym is not defined: ", $synonym->def()->text(); |
358
|
|
|
|
|
|
|
} |
359
|
|
|
|
|
|
|
|
360
|
5880
|
|
|
|
|
13934
|
push @result, $synonym->def_as_string().$s_as_string; |
361
|
|
|
|
|
|
|
} |
362
|
2141
|
|
|
|
|
8569
|
return @result; |
363
|
|
|
|
|
|
|
} |
364
|
|
|
|
|
|
|
|
365
|
|
|
|
|
|
|
=head2 xref_set |
366
|
|
|
|
|
|
|
|
367
|
|
|
|
|
|
|
Usage - $term->xref_set() or $term->xref_set($dbxref_set) |
368
|
|
|
|
|
|
|
Returns - a Dbxref set (OBO::Util::DbxrefSet) with the analogous xref(s) of this term in another vocabulary |
369
|
|
|
|
|
|
|
Args - a set of analogous xref(s) (OBO::Util::DbxrefSet) of this term in another vocabulary |
370
|
|
|
|
|
|
|
Function - gets/sets the analogous xref(s) set of this term in another vocabulary |
371
|
|
|
|
|
|
|
|
372
|
|
|
|
|
|
|
=cut |
373
|
|
|
|
|
|
|
|
374
|
|
|
|
|
|
|
sub xref_set { |
375
|
4580
|
100
|
|
4580
|
1
|
8811
|
$_[0]->{XREF_SET} = $_[1] if ($_[1]); |
376
|
4580
|
|
|
|
|
14367
|
return $_[0]->{XREF_SET}; |
377
|
|
|
|
|
|
|
} |
378
|
|
|
|
|
|
|
|
379
|
|
|
|
|
|
|
=head2 xref_set_as_string |
380
|
|
|
|
|
|
|
|
381
|
|
|
|
|
|
|
Usage - $term->xref_set_as_string() or $term->xref_set_as_string("[Reactome:20610, EC:2.3.2.12]") |
382
|
|
|
|
|
|
|
Returns - the dbxref set with the analogous xref(s) of this term; [] if the set is empty |
383
|
|
|
|
|
|
|
Args - the dbxref set with the analogous xref(s) of this term |
384
|
|
|
|
|
|
|
Function - gets/sets the dbxref set with the analogous xref(s) of this term |
385
|
|
|
|
|
|
|
Remark - make sure that colons (,) are scaped (\,) when necessary |
386
|
|
|
|
|
|
|
|
387
|
|
|
|
|
|
|
=cut |
388
|
|
|
|
|
|
|
|
389
|
|
|
|
|
|
|
sub xref_set_as_string { |
390
|
3876
|
|
|
3876
|
1
|
6368
|
my $xref_as_string = $_[1]; |
391
|
3876
|
100
|
|
|
|
7748
|
if ($xref_as_string) { |
392
|
1176
|
|
|
|
|
1933
|
my $xref_set = $_[0]->{XREF_SET}; |
393
|
|
|
|
|
|
|
|
394
|
1176
|
|
|
|
|
2272
|
my ($e, $entry) = __dbxref($xref_set, $xref_as_string); |
395
|
1176
|
50
|
|
|
|
2663
|
if ($e == -1) { |
396
|
0
|
|
|
|
|
0
|
croak "ERROR: Check the 'dbxref' field of '", $entry, "' (term ID = ", $_[0]->id(), ")." ; |
397
|
|
|
|
|
|
|
} |
398
|
|
|
|
|
|
|
|
399
|
1176
|
|
|
|
|
2261
|
$_[0]->{XREF_SET} = $xref_set; # We are overwriting the existing set; otherwise, add the new elements to the existing set! |
400
|
|
|
|
|
|
|
} |
401
|
3876
|
|
|
|
|
8398
|
my @result = $_[0]->xref_set()->get_set(); |
402
|
|
|
|
|
|
|
} |
403
|
|
|
|
|
|
|
|
404
|
|
|
|
|
|
|
=head2 property_value |
405
|
|
|
|
|
|
|
|
406
|
|
|
|
|
|
|
Usage - $term->property_value() or $term->property_value($p_value1, $p_value2, $p_value3, ...) |
407
|
|
|
|
|
|
|
Returns - an array with the property value(s) of this term |
408
|
|
|
|
|
|
|
Args - the relationship(s) (OBO::Core::Relationship) of this term with its property value(s) |
409
|
|
|
|
|
|
|
Function - gets/sets the property_value(s) of this term |
410
|
|
|
|
|
|
|
Remark - WARNING: this code might change! |
411
|
|
|
|
|
|
|
|
412
|
|
|
|
|
|
|
=cut |
413
|
|
|
|
|
|
|
|
414
|
|
|
|
|
|
|
sub property_value { |
415
|
|
|
|
|
|
|
# TODO WARNING: this code might change! |
416
|
2700
|
|
|
2700
|
1
|
4614
|
my ($self, @co) = @_; |
417
|
|
|
|
|
|
|
|
418
|
2700
|
|
|
|
|
4324
|
foreach my $i (@co) { |
419
|
7
|
|
|
|
|
24
|
$self->{PROPERTY_VALUE}->add($i); |
420
|
|
|
|
|
|
|
} |
421
|
2700
|
|
|
|
|
10480
|
return $self->{PROPERTY_VALUE}; |
422
|
|
|
|
|
|
|
} |
423
|
|
|
|
|
|
|
|
424
|
|
|
|
|
|
|
=head2 class_of |
425
|
|
|
|
|
|
|
|
426
|
|
|
|
|
|
|
Usage - $term->class_of() or $term->class_of($instance1, $instance2, $instance3, ...) |
427
|
|
|
|
|
|
|
Returns - an array with the instance(s) of this term |
428
|
|
|
|
|
|
|
Args - the instance(s) (OBO::Core::Instance) of this term |
429
|
|
|
|
|
|
|
Function - gets/sets the instance(s) of this term |
430
|
|
|
|
|
|
|
|
431
|
|
|
|
|
|
|
=cut |
432
|
|
|
|
|
|
|
|
433
|
|
|
|
|
|
|
sub class_of { |
434
|
123
|
|
|
123
|
1
|
209
|
my ($self, @co) = @_; |
435
|
|
|
|
|
|
|
|
436
|
123
|
|
|
|
|
232
|
foreach my $i (@co) { |
437
|
4
|
|
|
|
|
15
|
$self->{CLASS_OF}->add($i); |
438
|
4
|
|
|
|
|
15
|
$i->instance_of($self); # make the instance aware of its class (term) |
439
|
|
|
|
|
|
|
} |
440
|
123
|
|
|
|
|
506
|
return $self->{CLASS_OF}; |
441
|
|
|
|
|
|
|
} |
442
|
|
|
|
|
|
|
|
443
|
|
|
|
|
|
|
=head2 is_class_of |
444
|
|
|
|
|
|
|
|
445
|
|
|
|
|
|
|
Usage - $term->is_class_of($instance) |
446
|
|
|
|
|
|
|
Returns - either 1 (true) or 0 (false) |
447
|
|
|
|
|
|
|
Args - an instance (OBO::Core::Instance) of which this object might be class of |
448
|
|
|
|
|
|
|
Function - tells whether this object is a class of $instance |
449
|
|
|
|
|
|
|
|
450
|
|
|
|
|
|
|
=cut |
451
|
|
|
|
|
|
|
|
452
|
|
|
|
|
|
|
sub is_class_of { |
453
|
12
|
|
66
|
12
|
1
|
102
|
return (defined $_[1] && $_[0]->{CLASS_OF}->contains($_[1])); |
454
|
|
|
|
|
|
|
} |
455
|
|
|
|
|
|
|
|
456
|
|
|
|
|
|
|
=head2 intersection_of |
457
|
|
|
|
|
|
|
|
458
|
|
|
|
|
|
|
Usage - $term->intersection_of() or $term->intersection_of($t1, $t2, $r1, ...) |
459
|
|
|
|
|
|
|
Returns - an array with the terms/relations which define this term |
460
|
|
|
|
|
|
|
Args - a set (strings) of terms/relations which define this term |
461
|
|
|
|
|
|
|
Function - gets/sets the set of terms/relationships defining this term |
462
|
|
|
|
|
|
|
|
463
|
|
|
|
|
|
|
=cut |
464
|
|
|
|
|
|
|
|
465
|
|
|
|
|
|
|
sub intersection_of { |
466
|
3389
|
|
|
3389
|
1
|
4724
|
my $self = shift; |
467
|
3389
|
50
|
|
|
|
9817
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
468
|
0
|
|
|
|
|
0
|
$self->{INTERSECTION_OF}->add_all(@_); |
469
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
470
|
4
|
|
|
|
|
14
|
$self->{INTERSECTION_OF}->add(shift); |
471
|
|
|
|
|
|
|
} |
472
|
3389
|
|
|
|
|
11231
|
return $self->{INTERSECTION_OF}->get_set(); |
473
|
|
|
|
|
|
|
} |
474
|
|
|
|
|
|
|
|
475
|
|
|
|
|
|
|
=head2 union_of |
476
|
|
|
|
|
|
|
|
477
|
|
|
|
|
|
|
Usage - $term->union_of() or $term->union_of($t1, $t2, $r1, ...) |
478
|
|
|
|
|
|
|
Returns - an array with the terms/relations which define this term |
479
|
|
|
|
|
|
|
Args - a set (strings) of terms/relations which define this term |
480
|
|
|
|
|
|
|
Function - gets/sets the set of terms/relationships defining this term |
481
|
|
|
|
|
|
|
|
482
|
|
|
|
|
|
|
=cut |
483
|
|
|
|
|
|
|
|
484
|
|
|
|
|
|
|
sub union_of { |
485
|
3387
|
|
|
3387
|
1
|
4158
|
my $self = shift; |
486
|
3387
|
50
|
|
|
|
8906
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
487
|
0
|
|
|
|
|
0
|
$self->{UNION_OF}->add_all(@_); |
488
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
489
|
2
|
|
|
|
|
8
|
$self->{UNION_OF}->add(shift); |
490
|
|
|
|
|
|
|
} |
491
|
3387
|
|
|
|
|
9314
|
return $self->{UNION_OF}->get_set(); |
492
|
|
|
|
|
|
|
} |
493
|
|
|
|
|
|
|
|
494
|
|
|
|
|
|
|
=head2 disjoint_from |
495
|
|
|
|
|
|
|
|
496
|
|
|
|
|
|
|
Usage - $term->disjoint_from() or $term->disjoint_from($disjoint_term_id1, $disjoint_term_id2, $disjoint_term_id3, ...) |
497
|
|
|
|
|
|
|
Returns - the disjoint term id(s) (string(s)) from this one |
498
|
|
|
|
|
|
|
Args - the term id(s) (string) that is (are) disjoint from this one |
499
|
|
|
|
|
|
|
Function - gets/sets the disjoint term(s) from this one |
500
|
|
|
|
|
|
|
|
501
|
|
|
|
|
|
|
=cut |
502
|
|
|
|
|
|
|
|
503
|
|
|
|
|
|
|
sub disjoint_from { |
504
|
3402
|
|
|
3402
|
1
|
4469
|
my $self = shift; |
505
|
3402
|
100
|
|
|
|
8640
|
if (scalar(@_) > 1) { |
|
|
100
|
|
|
|
|
|
506
|
1
|
|
|
|
|
5
|
$self->{DISJOINT_FROM}->add_all(@_); |
507
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
508
|
15
|
|
|
|
|
50
|
$self->{DISJOINT_FROM}->add(shift); |
509
|
|
|
|
|
|
|
} |
510
|
3402
|
|
|
|
|
9968
|
return $self->{DISJOINT_FROM}->get_set(); |
511
|
|
|
|
|
|
|
} |
512
|
|
|
|
|
|
|
|
513
|
|
|
|
|
|
|
=head2 created_by |
514
|
|
|
|
|
|
|
|
515
|
|
|
|
|
|
|
Usage - print $term->created_by() or $term->created_by("erick_antezana") |
516
|
|
|
|
|
|
|
Returns - name (string) of the creator of the term, may be a short username, initials or ID |
517
|
|
|
|
|
|
|
Args - name (string) of the creator of the term, may be a short username, initials or ID |
518
|
|
|
|
|
|
|
Function - gets/sets the name of the creator of the term |
519
|
|
|
|
|
|
|
|
520
|
|
|
|
|
|
|
=cut |
521
|
|
|
|
|
|
|
|
522
|
|
|
|
|
|
|
sub created_by { |
523
|
2717
|
100
|
|
2717
|
1
|
6308
|
$_[0]->{CREATED_BY} = $_[1] if ($_[1]); |
524
|
2717
|
|
|
|
|
8307
|
return $_[0]->{CREATED_BY}; |
525
|
|
|
|
|
|
|
} |
526
|
|
|
|
|
|
|
|
527
|
|
|
|
|
|
|
=head2 creation_date |
528
|
|
|
|
|
|
|
|
529
|
|
|
|
|
|
|
Usage - print $term->creation_date() or $term->creation_date("2010-04-13T01:32:36Z") |
530
|
|
|
|
|
|
|
Returns - date (string) of creation of the term specified in ISO 8601 format |
531
|
|
|
|
|
|
|
Args - date (string) of creation of the term specified in ISO 8601 format |
532
|
|
|
|
|
|
|
Function - gets/sets the date of creation of the term |
533
|
|
|
|
|
|
|
Remark - You can get an ISO 8601 date as follows: |
534
|
|
|
|
|
|
|
use POSIX qw(strftime); |
535
|
|
|
|
|
|
|
my $datetime = strftime("%Y-%m-%dT%H:%M:%S", localtime()); |
536
|
|
|
|
|
|
|
|
537
|
|
|
|
|
|
|
=cut |
538
|
|
|
|
|
|
|
|
539
|
|
|
|
|
|
|
sub creation_date { |
540
|
2719
|
100
|
|
2719
|
1
|
5310
|
$_[0]->{CREATION_DATE} = $_[1] if ($_[1]); |
541
|
2719
|
|
|
|
|
7885
|
return $_[0]->{CREATION_DATE}; |
542
|
|
|
|
|
|
|
} |
543
|
|
|
|
|
|
|
|
544
|
|
|
|
|
|
|
=head2 modified_by |
545
|
|
|
|
|
|
|
|
546
|
|
|
|
|
|
|
Usage - print $term->modified_by() or $term->modified_by("erick_antezana") |
547
|
|
|
|
|
|
|
Returns - name (string) of the modificator of the term, may be a short username, initials or ID |
548
|
|
|
|
|
|
|
Args - name (string) of the modificator of the term, may be a short username, initials or ID |
549
|
|
|
|
|
|
|
Function - gets/sets the name of the modificator of the term |
550
|
|
|
|
|
|
|
|
551
|
|
|
|
|
|
|
=cut |
552
|
|
|
|
|
|
|
|
553
|
|
|
|
|
|
|
sub modified_by { |
554
|
|
|
|
|
|
|
# TODO WARNING: This is not going to be in the OBO spec. Use property_values instead... |
555
|
2693
|
100
|
|
2693
|
1
|
5288
|
$_[0]->{MODIFIED_BY} = $_[1] if ($_[1]); |
556
|
2693
|
|
|
|
|
7208
|
return $_[0]->{MODIFIED_BY}; |
557
|
|
|
|
|
|
|
} |
558
|
|
|
|
|
|
|
|
559
|
|
|
|
|
|
|
=head2 modification_date |
560
|
|
|
|
|
|
|
|
561
|
|
|
|
|
|
|
Usage - print $term->modification_date() or $term->modification_date("2010-04-13T01:32:36Z") |
562
|
|
|
|
|
|
|
Returns - date (string) of modification of the term specified in ISO 8601 format |
563
|
|
|
|
|
|
|
Args - date (string) of modification of the term specified in ISO 8601 format |
564
|
|
|
|
|
|
|
Function - gets/sets the date of modification of the term |
565
|
|
|
|
|
|
|
Remark - You can get an ISO 8601 date as follows: |
566
|
|
|
|
|
|
|
use POSIX qw(strftime); |
567
|
|
|
|
|
|
|
my $datetime = strftime("%Y-%m-%dT%H:%M:%S", localtime()); |
568
|
|
|
|
|
|
|
|
569
|
|
|
|
|
|
|
=cut |
570
|
|
|
|
|
|
|
|
571
|
|
|
|
|
|
|
sub modification_date { |
572
|
|
|
|
|
|
|
# TODO WARNING: This is not going to be in the OBO spec. Use property_values instead... |
573
|
2693
|
100
|
|
2693
|
1
|
4975
|
$_[0]->{MODIFICATION_DATE} = $_[1] if ($_[1]); |
574
|
2693
|
|
|
|
|
7788
|
return $_[0]->{MODIFICATION_DATE}; |
575
|
|
|
|
|
|
|
} |
576
|
|
|
|
|
|
|
|
577
|
|
|
|
|
|
|
=head2 is_obsolete |
578
|
|
|
|
|
|
|
|
579
|
|
|
|
|
|
|
Usage - $term->is_obsolete(1) or print $term->is_obsolete() |
580
|
|
|
|
|
|
|
Returns - either 1 (true) or 0 (false) |
581
|
|
|
|
|
|
|
Args - either 1 (true) or 0 (false) |
582
|
|
|
|
|
|
|
Function - tells whether the term is obsolete or not. 'false' by default. |
583
|
|
|
|
|
|
|
|
584
|
|
|
|
|
|
|
=cut |
585
|
|
|
|
|
|
|
|
586
|
|
|
|
|
|
|
sub is_obsolete { |
587
|
3421
|
50
|
66
|
3421
|
1
|
7212
|
if (defined $_[1] && ($_[1] == 1 || $_[1] == 0)) { $_[0]->{IS_OBSOLETE} = $_[1] } |
|
31
|
|
66
|
|
|
63
|
|
588
|
3421
|
100
|
66
|
|
|
15865
|
return ($_[0]->{IS_OBSOLETE} && $_[0]->{IS_OBSOLETE} == 1)?1:0; |
589
|
|
|
|
|
|
|
} |
590
|
|
|
|
|
|
|
|
591
|
|
|
|
|
|
|
=head2 replaced_by |
592
|
|
|
|
|
|
|
|
593
|
|
|
|
|
|
|
Usage - $term->replaced_by() or $term->replaced_by($id1, $id2, $id3, ...) |
594
|
|
|
|
|
|
|
Returns - a set (OBO::Util::Set) with the id(s) of the replacing term(s) |
595
|
|
|
|
|
|
|
Args - the the id(s) of the replacing term(s) (string) |
596
|
|
|
|
|
|
|
Function - gets/sets the the id(s) of the replacing term(s) |
597
|
|
|
|
|
|
|
|
598
|
|
|
|
|
|
|
=cut |
599
|
|
|
|
|
|
|
|
600
|
|
|
|
|
|
|
sub replaced_by { |
601
|
3385
|
|
|
3385
|
1
|
4355
|
my $self = shift; |
602
|
3385
|
50
|
|
|
|
9328
|
if (scalar(@_) > 1) { |
|
|
50
|
|
|
|
|
|
603
|
0
|
|
|
|
|
0
|
$self->{REPLACED_BY}->add_all(@_); |
604
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
605
|
0
|
|
|
|
|
0
|
$self->{REPLACED_BY}->add(shift); |
606
|
|
|
|
|
|
|
} |
607
|
3385
|
|
|
|
|
12358
|
return $self->{REPLACED_BY}; |
608
|
|
|
|
|
|
|
} |
609
|
|
|
|
|
|
|
|
610
|
|
|
|
|
|
|
=head2 consider |
611
|
|
|
|
|
|
|
|
612
|
|
|
|
|
|
|
Usage - $term->consider() or $term->consider($id1, $id2, $id3, ...) |
613
|
|
|
|
|
|
|
Returns - a set (OBO::Util::Set) with the appropiate substitute(s) for an obsolete term |
614
|
|
|
|
|
|
|
Args - the appropiate substitute(s) for an obsolete term (string) |
615
|
|
|
|
|
|
|
Function - gets/sets the appropiate substitute(s) for this obsolete term |
616
|
|
|
|
|
|
|
|
617
|
|
|
|
|
|
|
=cut |
618
|
|
|
|
|
|
|
|
619
|
|
|
|
|
|
|
sub consider { |
620
|
3385
|
|
|
3385
|
1
|
4287
|
my $self = shift; |
621
|
3385
|
50
|
|
|
|
9096
|
if (scalar(@_) > 1) { |
|
|
50
|
|
|
|
|
|
622
|
0
|
|
|
|
|
0
|
$self->{CONSIDER}->add_all(@_); |
623
|
|
|
|
|
|
|
} elsif (scalar(@_) == 1) { |
624
|
0
|
|
|
|
|
0
|
$self->{CONSIDER}->add(shift); |
625
|
|
|
|
|
|
|
} |
626
|
3385
|
|
|
|
|
10160
|
return $self->{CONSIDER}; |
627
|
|
|
|
|
|
|
} |
628
|
|
|
|
|
|
|
|
629
|
|
|
|
|
|
|
=head2 builtin |
630
|
|
|
|
|
|
|
|
631
|
|
|
|
|
|
|
Usage - $term->builtin() or $term->builtin(1) or $term->builtin(0) |
632
|
|
|
|
|
|
|
Returns - tells if this term is builtin to the OBO format; false by default |
633
|
|
|
|
|
|
|
Args - 1 (true) or 0 (false) |
634
|
|
|
|
|
|
|
Function - gets/sets the value indicating whether this term is builtin to the OBO format |
635
|
|
|
|
|
|
|
|
636
|
|
|
|
|
|
|
=cut |
637
|
|
|
|
|
|
|
|
638
|
|
|
|
|
|
|
sub builtin { |
639
|
2746
|
50
|
33
|
2746
|
1
|
5980
|
if (defined $_[1] && ($_[1] == 1 || $_[1] == 0)) { $_[0]->{BUILTIN} = $_[1] } |
|
8
|
|
66
|
|
|
11
|
|
640
|
2746
|
50
|
33
|
|
|
12000
|
return ($_[0]->{BUILTIN} && $_[0]->{BUILTIN} == 1)?1:0; |
641
|
|
|
|
|
|
|
} |
642
|
|
|
|
|
|
|
|
643
|
|
|
|
|
|
|
=head2 equals |
644
|
|
|
|
|
|
|
|
645
|
|
|
|
|
|
|
Usage - print $term->equals($another_term) |
646
|
|
|
|
|
|
|
Returns - either 1 (true) or 0 (false) |
647
|
|
|
|
|
|
|
Args - the term (OBO::Core::Term) to compare with |
648
|
|
|
|
|
|
|
Function - tells whether this term is equal to the parameter |
649
|
|
|
|
|
|
|
|
650
|
|
|
|
|
|
|
=cut |
651
|
|
|
|
|
|
|
|
652
|
|
|
|
|
|
|
sub equals { |
653
|
34
|
50
|
33
|
34
|
1
|
164
|
if ($_[1] && eval { $_[1]->isa('OBO::Core::Term') }) { |
|
34
|
|
|
|
|
251
|
|
654
|
34
|
50
|
33
|
|
|
326
|
return (defined $_[1] && $_[0]->{'ID'} eq $_[1]->{'ID'})?1:0; |
655
|
|
|
|
|
|
|
} else { |
656
|
0
|
|
|
|
|
0
|
croak "An unrecognized object type (not a OBO::Core::Term) was found: '", $_[1], "'"; |
657
|
|
|
|
|
|
|
} |
658
|
|
|
|
|
|
|
} |
659
|
|
|
|
|
|
|
|
660
|
|
|
|
|
|
|
sub __dbxref () { |
661
|
1184
|
50
|
|
1184
|
|
2778
|
caller eq __PACKAGE__ or croak "You cannot call this (__dbxref) prived method!"; |
662
|
|
|
|
|
|
|
# |
663
|
|
|
|
|
|
|
# $_[0] ==> set |
664
|
|
|
|
|
|
|
# $_[1] ==> dbxref string |
665
|
|
|
|
|
|
|
# |
666
|
1184
|
|
|
|
|
1564
|
my $dbxref_set = $_[0]; |
667
|
1184
|
|
|
|
|
1503
|
my $dbxref_as_string = $_[1]; |
668
|
|
|
|
|
|
|
|
669
|
1184
|
|
|
|
|
1830
|
$dbxref_as_string =~ s/^\[//; |
670
|
1184
|
|
|
|
|
1823
|
$dbxref_as_string =~ s/\]$//; |
671
|
1184
|
|
|
|
|
1720
|
$dbxref_as_string =~ s/\\,/;;;;/g; # trick to keep the comma's |
672
|
1184
|
|
|
|
|
1694
|
$dbxref_as_string =~ s/\\"/;;;;;/g; # trick to keep the double quote's |
673
|
|
|
|
|
|
|
|
674
|
1184
|
|
|
|
|
2274
|
my @lineas = $dbxref_as_string =~ /\"([^\"]*)\"/g; # get the double-quoted pieces |
675
|
1184
|
|
|
|
|
2111
|
foreach my $l (@lineas) { |
676
|
8
|
|
|
|
|
14
|
my $cp = $l; |
677
|
8
|
|
|
|
|
16
|
$l =~ s/,/;;;;/g; # trick to keep the comma's |
678
|
8
|
|
|
|
|
135
|
$dbxref_as_string =~ s/\Q$cp\E/$l/; |
679
|
|
|
|
|
|
|
} |
680
|
|
|
|
|
|
|
|
681
|
1184
|
|
|
|
|
3662
|
my $r_db_acc = qr/([ \*\.\w-]*):([ ;'\#~\w:\\\+\?\{\}\$\/\(\)\[\]\.=&!%_-]*)/o; |
682
|
1184
|
|
|
|
|
3086
|
my $r_desc = qr/\s+\"([^\"]*)\"/o; |
683
|
1184
|
|
|
|
|
2900
|
my $r_mod = qr/\s+(\{[\w ]+=[\w ]+\})/o; |
684
|
|
|
|
|
|
|
|
685
|
1184
|
|
|
|
|
3043
|
my @dbxrefs = split (',', $dbxref_as_string); |
686
|
|
|
|
|
|
|
|
687
|
1184
|
|
|
|
|
2097
|
foreach my $entry (@dbxrefs) { |
688
|
1192
|
|
|
|
|
2017
|
my ($match, $db, $acc, $desc, $mod) = undef; |
689
|
1192
|
|
|
|
|
3613
|
my $dbxref = OBO::Core::Dbxref->new(); |
690
|
1192
|
100
|
|
|
|
30581
|
if ($entry =~ m/$r_db_acc$r_desc$r_mod?/) { |
|
|
50
|
|
|
|
|
|
691
|
8
|
|
|
|
|
21
|
$db = __unescape($1); |
692
|
8
|
|
|
|
|
24
|
$acc = __unescape($2); |
693
|
8
|
|
|
|
|
22
|
$desc = __unescape($3); |
694
|
8
|
100
|
|
|
|
48
|
$mod = __unescape($4) if ($4); |
695
|
|
|
|
|
|
|
} elsif ($entry =~ m/$r_db_acc$r_desc?$r_mod?/) { |
696
|
1184
|
|
|
|
|
2423
|
$db = __unescape($1); |
697
|
1184
|
|
|
|
|
2403
|
$acc = __unescape($2); |
698
|
1184
|
50
|
|
|
|
3217
|
$desc = __unescape($3) if ($3); |
699
|
1184
|
50
|
|
|
|
2803
|
$mod = __unescape($4) if ($4); |
700
|
|
|
|
|
|
|
} else { |
701
|
0
|
|
|
|
|
0
|
return (-1, $entry); |
702
|
|
|
|
|
|
|
} |
703
|
|
|
|
|
|
|
|
704
|
|
|
|
|
|
|
# set the dbxref: |
705
|
1192
|
|
|
|
|
5376
|
$dbxref->name($db.':'.$acc); |
706
|
1192
|
100
|
|
|
|
3033
|
$dbxref->description($desc) if (defined $desc); |
707
|
1192
|
100
|
|
|
|
2393
|
$dbxref->modifier($mod) if (defined $mod); |
708
|
1192
|
|
|
|
|
3457
|
$dbxref_set->add($dbxref); |
709
|
|
|
|
|
|
|
} |
710
|
1184
|
|
|
|
|
4097
|
return 1; |
711
|
|
|
|
|
|
|
} |
712
|
|
|
|
|
|
|
|
713
|
|
|
|
|
|
|
sub __unescape { |
714
|
2395
|
50
|
|
2395
|
|
5298
|
caller eq __PACKAGE__ or die; |
715
|
2395
|
|
|
|
|
4363
|
my $match = $_[0]; |
716
|
2395
|
|
|
|
|
3106
|
$match =~ s/;;;;;/\\"/g; |
717
|
2395
|
|
|
|
|
3111
|
$match =~ s/;;;;/\\,/g; |
718
|
2395
|
|
|
|
|
4836
|
return $match; |
719
|
|
|
|
|
|
|
} |
720
|
|
|
|
|
|
|
|
721
|
|
|
|
|
|
|
1; |
722
|
|
|
|
|
|
|
|
723
|
|
|
|
|
|
|
__END__ |