| line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
|
1
|
|
|
|
|
|
|
package Bio::SeqFeature::Lite; |
|
2
|
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
=head1 NAME |
|
4
|
|
|
|
|
|
|
|
|
5
|
|
|
|
|
|
|
Bio::SeqFeature::Lite - Lightweight Bio::SeqFeatureI class |
|
6
|
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
=head1 SYNOPSIS |
|
8
|
|
|
|
|
|
|
|
|
9
|
|
|
|
|
|
|
# create a simple feature with no internal structure |
|
10
|
|
|
|
|
|
|
$f = Bio::SeqFeature::Lite->new(-start => 1000, |
|
11
|
|
|
|
|
|
|
-stop => 2000, |
|
12
|
|
|
|
|
|
|
-type => 'transcript', |
|
13
|
|
|
|
|
|
|
-name => 'alpha-1 antitrypsin', |
|
14
|
|
|
|
|
|
|
-desc => 'an enzyme inhibitor', |
|
15
|
|
|
|
|
|
|
); |
|
16
|
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
# create a feature composed of multiple segments, all of type "similarity" |
|
18
|
|
|
|
|
|
|
$f = Bio::SeqFeature::Lite->new(-segments => [[1000,1100],[1500,1550],[1800,2000]], |
|
19
|
|
|
|
|
|
|
-name => 'ABC-3', |
|
20
|
|
|
|
|
|
|
-type => 'gapped_alignment', |
|
21
|
|
|
|
|
|
|
-subtype => 'similarity'); |
|
22
|
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
# build up a gene exon by exon |
|
24
|
|
|
|
|
|
|
$e1 = Bio::SeqFeature::Lite->new(-start=>1,-stop=>100,-type=>'exon'); |
|
25
|
|
|
|
|
|
|
$e2 = Bio::SeqFeature::Lite->new(-start=>150,-stop=>200,-type=>'exon'); |
|
26
|
|
|
|
|
|
|
$e3 = Bio::SeqFeature::Lite->new(-start=>300,-stop=>500,-type=>'exon'); |
|
27
|
|
|
|
|
|
|
$f = Bio::SeqFeature::Lite->new(-segments=>[$e1,$e2,$e3],-type=>'gene'); |
|
28
|
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
=head1 DESCRIPTION |
|
30
|
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
This is a simple Bio::SeqFeatureI-compliant object that is compatible |
|
32
|
|
|
|
|
|
|
with Bio::Graphics::Panel. With it you can create lightweight feature |
|
33
|
|
|
|
|
|
|
objects for drawing. |
|
34
|
|
|
|
|
|
|
|
|
35
|
|
|
|
|
|
|
All methods are as described in L with the following additions: |
|
36
|
|
|
|
|
|
|
|
|
37
|
|
|
|
|
|
|
=head2 The new() Constructor |
|
38
|
|
|
|
|
|
|
|
|
39
|
|
|
|
|
|
|
$feature = Bio::SeqFeature::Lite->new(@args); |
|
40
|
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
This method creates a new feature object. You can create a simple |
|
42
|
|
|
|
|
|
|
feature that contains no subfeatures, or a hierarchically nested object. |
|
43
|
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
Arguments are as follows: |
|
45
|
|
|
|
|
|
|
|
|
46
|
|
|
|
|
|
|
-seq_id the reference sequence |
|
47
|
|
|
|
|
|
|
-start the start position of the feature |
|
48
|
|
|
|
|
|
|
-end the stop position of the feature |
|
49
|
|
|
|
|
|
|
-stop an alias for end |
|
50
|
|
|
|
|
|
|
-name the feature name (returned by seqname()) |
|
51
|
|
|
|
|
|
|
-type the feature type (returned by primary_tag()) |
|
52
|
|
|
|
|
|
|
-primary_tag the same as -type |
|
53
|
|
|
|
|
|
|
-source the source tag |
|
54
|
|
|
|
|
|
|
-score the feature score (for GFF compatibility) |
|
55
|
|
|
|
|
|
|
-desc a description of the feature |
|
56
|
|
|
|
|
|
|
-segments a list of subfeatures (see below) |
|
57
|
|
|
|
|
|
|
-subtype the type to use when creating subfeatures |
|
58
|
|
|
|
|
|
|
-strand the strand of the feature (one of -1, 0 or +1) |
|
59
|
|
|
|
|
|
|
-phase the phase of the feature (0..2) |
|
60
|
|
|
|
|
|
|
-seq a dna or protein sequence string to attach to feature |
|
61
|
|
|
|
|
|
|
-id an alias for -name |
|
62
|
|
|
|
|
|
|
-seqname an alias for -name |
|
63
|
|
|
|
|
|
|
-display_id an alias for -name |
|
64
|
|
|
|
|
|
|
-display_name an alias for -name (do you get the idea the API has changed?) |
|
65
|
|
|
|
|
|
|
-primary_id unique database ID |
|
66
|
|
|
|
|
|
|
-url a URL to link to when rendered with Bio::Graphics |
|
67
|
|
|
|
|
|
|
-attributes a hashref of tag value attributes, in which the key is the tag |
|
68
|
|
|
|
|
|
|
and the value is an array reference of values |
|
69
|
|
|
|
|
|
|
-factory a reference to a feature factory, used for compatibility with |
|
70
|
|
|
|
|
|
|
more obscure parts of Bio::DB::GFF |
|
71
|
|
|
|
|
|
|
|
|
72
|
|
|
|
|
|
|
The subfeatures passed in -segments may be an array of |
|
73
|
|
|
|
|
|
|
Bio::SeqFeature::Lite objects, or an array of [$start,$stop] |
|
74
|
|
|
|
|
|
|
pairs. Each pair should be a two-element array reference. In the |
|
75
|
|
|
|
|
|
|
latter case, the feature type passed in -subtype will be used when |
|
76
|
|
|
|
|
|
|
creating the subfeatures. |
|
77
|
|
|
|
|
|
|
|
|
78
|
|
|
|
|
|
|
If no feature type is passed, then it defaults to "feature". |
|
79
|
|
|
|
|
|
|
|
|
80
|
|
|
|
|
|
|
=head2 Non-SeqFeatureI methods |
|
81
|
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
A number of new methods are provided for compatibility with |
|
83
|
|
|
|
|
|
|
Ace::Sequence, which has a slightly different API from SeqFeatureI: |
|
84
|
|
|
|
|
|
|
|
|
85
|
|
|
|
|
|
|
=over 4 |
|
86
|
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
=item url() |
|
88
|
|
|
|
|
|
|
|
|
89
|
|
|
|
|
|
|
Get/set the URL that the graphical rendering of this feature will link to. |
|
90
|
|
|
|
|
|
|
|
|
91
|
|
|
|
|
|
|
=item add_segment(@segments) |
|
92
|
|
|
|
|
|
|
|
|
93
|
|
|
|
|
|
|
Add one or more segments (a subfeature). Segments can either be |
|
94
|
|
|
|
|
|
|
Feature objects, or [start,stop] arrays, as in the -segments argument |
|
95
|
|
|
|
|
|
|
to new(). The feature endpoints are automatically adjusted. |
|
96
|
|
|
|
|
|
|
|
|
97
|
|
|
|
|
|
|
=item segments() |
|
98
|
|
|
|
|
|
|
|
|
99
|
|
|
|
|
|
|
An alias for sub_SeqFeature(). |
|
100
|
|
|
|
|
|
|
|
|
101
|
|
|
|
|
|
|
=item get_SeqFeatures() |
|
102
|
|
|
|
|
|
|
|
|
103
|
|
|
|
|
|
|
Alias for sub_SeqFeature() |
|
104
|
|
|
|
|
|
|
|
|
105
|
|
|
|
|
|
|
=item get_all_SeqFeatures() |
|
106
|
|
|
|
|
|
|
|
|
107
|
|
|
|
|
|
|
Alias for sub_SeqFeature() |
|
108
|
|
|
|
|
|
|
|
|
109
|
|
|
|
|
|
|
=item merged_segments() |
|
110
|
|
|
|
|
|
|
|
|
111
|
|
|
|
|
|
|
Another alias for sub_SeqFeature(). |
|
112
|
|
|
|
|
|
|
|
|
113
|
|
|
|
|
|
|
=item stop() |
|
114
|
|
|
|
|
|
|
|
|
115
|
|
|
|
|
|
|
An alias for end(). |
|
116
|
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
=item name() |
|
118
|
|
|
|
|
|
|
|
|
119
|
|
|
|
|
|
|
An alias for seqname(). |
|
120
|
|
|
|
|
|
|
|
|
121
|
|
|
|
|
|
|
=item exons() |
|
122
|
|
|
|
|
|
|
|
|
123
|
|
|
|
|
|
|
An alias for sub_SeqFeature() (you don't want to know why!) |
|
124
|
|
|
|
|
|
|
|
|
125
|
|
|
|
|
|
|
=back |
|
126
|
|
|
|
|
|
|
|
|
127
|
|
|
|
|
|
|
=cut |
|
128
|
|
|
|
|
|
|
|
|
129
|
1
|
|
|
1
|
|
1075
|
use strict; |
|
|
1
|
|
|
|
|
1
|
|
|
|
1
|
|
|
|
|
25
|
|
|
130
|
|
|
|
|
|
|
|
|
131
|
1
|
|
|
1
|
|
3
|
use base qw(Bio::Root::Root Bio::SeqFeatureI Bio::LocationI Bio::SeqI); |
|
|
1
|
|
|
|
|
1
|
|
|
|
1
|
|
|
|
|
3034
|
|
|
132
|
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
*stop = \&end; |
|
134
|
|
|
|
|
|
|
*info = \&name; |
|
135
|
|
|
|
|
|
|
*seqname = \&name; |
|
136
|
|
|
|
|
|
|
*exons = *sub_SeqFeature = *merged_segments = \&segments; |
|
137
|
|
|
|
|
|
|
*get_all_SeqFeatures = *get_SeqFeatures = \&segments; |
|
138
|
|
|
|
|
|
|
*method = \&primary_tag; |
|
139
|
|
|
|
|
|
|
*source = \&source_tag; |
|
140
|
|
|
|
|
|
|
*get_tag_values = \&each_tag_value; |
|
141
|
|
|
|
|
|
|
*add_SeqFeature = \&add_segment; |
|
142
|
|
|
|
|
|
|
*get_all_tags = \&all_tags; |
|
143
|
|
|
|
|
|
|
*abs_ref = \&ref; |
|
144
|
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
# implement Bio::SeqI and FeatureHolderI interface |
|
146
|
|
|
|
|
|
|
|
|
147
|
0
|
|
|
0
|
1
|
0
|
sub primary_seq { return $_[0] } |
|
148
|
|
|
|
|
|
|
sub annotation { |
|
149
|
0
|
|
|
0
|
1
|
0
|
my ($obj,$value) = @_; |
|
150
|
0
|
0
|
|
|
|
0
|
if( defined $value ) { |
|
|
|
0
|
|
|
|
|
|
|
151
|
0
|
0
|
|
|
|
0
|
$obj->throw("object of class ".ref($value)." does not implement ". |
|
152
|
|
|
|
|
|
|
"Bio::AnnotationCollectionI. Too bad.") |
|
153
|
|
|
|
|
|
|
unless $value->isa("Bio::AnnotationCollectionI"); |
|
154
|
0
|
|
|
|
|
0
|
$obj->{'_annotation'} = $value; |
|
155
|
|
|
|
|
|
|
} elsif( ! defined $obj->{'_annotation'}) { |
|
156
|
0
|
|
|
|
|
0
|
$obj->{'_annotation'} = Bio::Annotation::Collection->new(); |
|
157
|
|
|
|
|
|
|
} |
|
158
|
0
|
|
|
|
|
0
|
return $obj->{'_annotation'}; |
|
159
|
|
|
|
|
|
|
} |
|
160
|
|
|
|
|
|
|
sub species { |
|
161
|
0
|
|
|
0
|
1
|
0
|
my ($self, $species) = @_; |
|
162
|
0
|
0
|
|
|
|
0
|
if ($species) { |
|
163
|
0
|
|
|
|
|
0
|
$self->{'species'} = $species; |
|
164
|
|
|
|
|
|
|
} else { |
|
165
|
0
|
|
|
|
|
0
|
return $self->{'species'}; |
|
166
|
|
|
|
|
|
|
} |
|
167
|
|
|
|
|
|
|
} |
|
168
|
0
|
|
|
0
|
1
|
0
|
sub is_remote { return } |
|
169
|
0
|
0
|
|
0
|
1
|
0
|
sub feature_count { return scalar @{shift->{segments} || []} } |
|
|
0
|
|
|
|
|
0
|
|
|
170
|
|
|
|
|
|
|
|
|
171
|
0
|
|
|
0
|
0
|
0
|
sub target { return; } |
|
172
|
0
|
|
|
0
|
0
|
0
|
sub hit { shift->target } |
|
173
|
|
|
|
|
|
|
|
|
174
|
|
|
|
|
|
|
sub type { |
|
175
|
0
|
|
|
0
|
0
|
0
|
my $self = shift; |
|
176
|
0
|
|
|
|
|
0
|
my $method = $self->primary_tag; |
|
177
|
0
|
|
|
|
|
0
|
my $source = $self->source_tag; |
|
178
|
0
|
0
|
|
|
|
0
|
return $source ne '' ? "$method:$source" : $method; |
|
179
|
|
|
|
|
|
|
} |
|
180
|
|
|
|
|
|
|
|
|
181
|
|
|
|
|
|
|
# usage: |
|
182
|
|
|
|
|
|
|
# Bio::SeqFeature::Lite->new( |
|
183
|
|
|
|
|
|
|
# -start => 1, |
|
184
|
|
|
|
|
|
|
# -end => 100, |
|
185
|
|
|
|
|
|
|
# -name => 'fred feature', |
|
186
|
|
|
|
|
|
|
# -strand => +1); |
|
187
|
|
|
|
|
|
|
# |
|
188
|
|
|
|
|
|
|
# Alternatively, use -segments => [ [start,stop],[start,stop]...] |
|
189
|
|
|
|
|
|
|
# to create a multisegmented feature. |
|
190
|
|
|
|
|
|
|
sub new { |
|
191
|
3
|
|
|
3
|
1
|
9
|
my $class= shift; |
|
192
|
3
|
50
|
|
|
|
5
|
$class = ref($class) if ref $class; |
|
193
|
3
|
|
|
|
|
7
|
my %arg = @_; |
|
194
|
|
|
|
|
|
|
|
|
195
|
3
|
|
|
|
|
5
|
my $self = bless {},$class; |
|
196
|
|
|
|
|
|
|
|
|
197
|
3
|
|
50
|
|
|
7
|
$arg{-strand} ||= 0; |
|
198
|
3
|
50
|
|
|
|
11
|
if ($arg{-strand} =~ /^[\+\-\.]$/){ |
|
199
|
3
|
100
|
|
|
|
16
|
($arg{-strand} eq "+") && ($self->{strand} = '1'); |
|
200
|
3
|
100
|
|
|
|
7
|
($arg{-strand} eq "-") && ($self->{strand} = '-1'); |
|
201
|
3
|
100
|
|
|
|
7
|
($arg{-strand} eq ".") && ($self->{strand} = '0'); |
|
202
|
|
|
|
|
|
|
} else { |
|
203
|
0
|
0
|
|
|
|
0
|
$self->{strand} = $arg{-strand} ? ($arg{-strand} >= 0 ? +1 : -1) : 0; |
|
|
|
0
|
|
|
|
|
|
|
204
|
|
|
|
|
|
|
} |
|
205
|
|
|
|
|
|
|
$self->{name} = $arg{-name} || $arg{-seqname} || $arg{-display_id} |
|
206
|
3
|
|
33
|
|
|
27
|
|| $arg{-display_name} || $arg{-id}; |
|
207
|
3
|
|
50
|
|
|
17
|
$self->{type} = $arg{-type} || $arg{-primary_tag} || 'feature'; |
|
208
|
3
|
50
|
|
|
|
7
|
$self->{subtype} = $arg{-subtype} if exists $arg{-subtype}; |
|
209
|
3
|
|
50
|
|
|
15
|
$self->{source} = $arg{-source} || $arg{-source_tag} || ''; |
|
210
|
3
|
50
|
|
|
|
4
|
$self->{score} = $arg{-score} if exists $arg{-score}; |
|
211
|
3
|
|
|
|
|
5
|
$self->{start} = $arg{-start}; |
|
212
|
3
|
100
|
|
|
|
8
|
$self->{stop} = exists $arg{-end} ? $arg{-end} : $arg{-stop}; |
|
213
|
3
|
|
33
|
|
|
9
|
$self->{ref} = $arg{-seq_id} || $arg{-ref}; |
|
214
|
3
|
|
33
|
|
|
9
|
$self->{attributes} = $arg{-attributes} || $arg{-tag}; |
|
215
|
3
|
|
|
|
|
4
|
for my $option (qw(class url seq phase desc primary_id)) { |
|
216
|
18
|
50
|
|
|
|
25
|
$self->{$option} = $arg{"-$option"} if exists $arg{"-$option"}; |
|
217
|
|
|
|
|
|
|
} |
|
218
|
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
# is_circular is needed for Bio::PrimarySeqI compliance |
|
220
|
3
|
|
50
|
|
|
10
|
$self->{is_circular} = $arg{-is_circular} || 0; |
|
221
|
|
|
|
|
|
|
|
|
222
|
|
|
|
|
|
|
# fix start, stop |
|
223
|
3
|
50
|
33
|
|
|
16
|
if (defined $self->{stop} && defined $self->{start} |
|
|
|
|
33
|
|
|
|
|
|
224
|
|
|
|
|
|
|
&& $self->{stop} < $self->{start}) { |
|
225
|
0
|
|
|
|
|
0
|
@{$self}{'start','stop'} = @{$self}{'stop','start'}; |
|
|
0
|
|
|
|
|
0
|
|
|
|
0
|
|
|
|
|
0
|
|
|
226
|
0
|
|
|
|
|
0
|
$self->{strand} *= -1; |
|
227
|
|
|
|
|
|
|
} |
|
228
|
|
|
|
|
|
|
|
|
229
|
3
|
|
|
|
|
2
|
my @segments; |
|
230
|
3
|
50
|
|
|
|
8
|
if (my $s = $arg{-segments}) { |
|
231
|
|
|
|
|
|
|
# NB: when $self ISA Bio::DB::SeqFeature the following invokes |
|
232
|
|
|
|
|
|
|
# Bio::DB::SeqFeature::add_segment and not |
|
233
|
|
|
|
|
|
|
# Bio::DB::SeqFeature::add_segment (as might be expected?) |
|
234
|
0
|
|
|
|
|
0
|
$self->add_segment(@$s); |
|
235
|
|
|
|
|
|
|
} |
|
236
|
|
|
|
|
|
|
|
|
237
|
3
|
|
|
|
|
6
|
$self; |
|
238
|
|
|
|
|
|
|
} |
|
239
|
|
|
|
|
|
|
|
|
240
|
|
|
|
|
|
|
sub add_segment { |
|
241
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
|
242
|
0
|
|
0
|
|
|
0
|
my $type = $self->{subtype} || $self->{type}; |
|
243
|
0
|
|
0
|
|
|
0
|
$self->{segments} ||= []; |
|
244
|
0
|
|
|
|
|
0
|
my $ref = $self->seq_id; |
|
245
|
0
|
|
|
|
|
0
|
my $name = $self->name; |
|
246
|
0
|
|
|
|
|
0
|
my $class = $self->class; |
|
247
|
0
|
|
|
|
|
0
|
my $source_tag = $self->source_tag; |
|
248
|
|
|
|
|
|
|
|
|
249
|
0
|
|
0
|
|
|
0
|
my $min_start = $self->start || 999_999_999_999; |
|
250
|
0
|
|
0
|
|
|
0
|
my $max_stop = $self->end || -999_999_999_999; |
|
251
|
|
|
|
|
|
|
|
|
252
|
0
|
|
|
|
|
0
|
my @segments = @{$self->{segments}}; |
|
|
0
|
|
|
|
|
0
|
|
|
253
|
|
|
|
|
|
|
|
|
254
|
0
|
|
|
|
|
0
|
for my $seg (@_) { |
|
255
|
0
|
0
|
|
|
|
0
|
if (ref($seg) eq 'ARRAY') { |
|
|
|
0
|
|
|
|
|
|
|
256
|
0
|
|
|
|
|
0
|
my ($start,$stop) = @{$seg}; |
|
|
0
|
|
|
|
|
0
|
|
|
257
|
0
|
0
|
0
|
|
|
0
|
next unless defined $start && defined $stop; # fixes an obscure bug somewhere above us |
|
258
|
0
|
|
|
|
|
0
|
my $strand = $self->{strand}; |
|
259
|
|
|
|
|
|
|
|
|
260
|
0
|
0
|
|
|
|
0
|
if ($start > $stop) { |
|
261
|
0
|
|
|
|
|
0
|
($start,$stop) = ($stop,$start); |
|
262
|
0
|
|
|
|
|
0
|
$strand = -1; |
|
263
|
|
|
|
|
|
|
} |
|
264
|
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
push @segments,$self->new(-start => $start, |
|
266
|
|
|
|
|
|
|
-stop => $stop, |
|
267
|
|
|
|
|
|
|
-strand => $strand, |
|
268
|
|
|
|
|
|
|
-ref => $ref, |
|
269
|
|
|
|
|
|
|
-type => $type, |
|
270
|
|
|
|
|
|
|
-name => $name, |
|
271
|
|
|
|
|
|
|
-class => $class, |
|
272
|
|
|
|
|
|
|
-phase => $self->{phase}, |
|
273
|
|
|
|
|
|
|
-score => $self->{score}, |
|
274
|
|
|
|
|
|
|
-source_tag => $source_tag, |
|
275
|
|
|
|
|
|
|
-attributes => $self->{attributes}, |
|
276
|
0
|
|
|
|
|
0
|
); |
|
277
|
0
|
0
|
|
|
|
0
|
$min_start = $start if $start < $min_start; |
|
278
|
0
|
0
|
|
|
|
0
|
$max_stop = $stop if $stop > $max_stop; |
|
279
|
|
|
|
|
|
|
|
|
280
|
|
|
|
|
|
|
} elsif (ref $seg) { |
|
281
|
0
|
|
|
|
|
0
|
push @segments,$seg; |
|
282
|
|
|
|
|
|
|
|
|
283
|
0
|
0
|
0
|
|
|
0
|
$min_start = $seg->start if ($seg->start && $seg->start < $min_start); |
|
284
|
0
|
0
|
0
|
|
|
0
|
$max_stop = $seg->end if ($seg->end && $seg->end > $max_stop); |
|
285
|
|
|
|
|
|
|
} |
|
286
|
|
|
|
|
|
|
} |
|
287
|
0
|
0
|
|
|
|
0
|
if (@segments) { |
|
288
|
0
|
|
|
|
|
0
|
local $^W = 0; # some warning of an uninitialized variable... |
|
289
|
0
|
|
|
|
|
0
|
$self->{segments} = \@segments; |
|
290
|
0
|
|
0
|
|
|
0
|
$self->{ref} ||= $self->{segments}[0]->seq_id; |
|
291
|
0
|
|
|
|
|
0
|
$self->{start} = $min_start; |
|
292
|
0
|
|
|
|
|
0
|
$self->{stop} = $max_stop; |
|
293
|
|
|
|
|
|
|
} |
|
294
|
|
|
|
|
|
|
} |
|
295
|
|
|
|
|
|
|
|
|
296
|
|
|
|
|
|
|
sub segments { |
|
297
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
|
298
|
0
|
0
|
|
|
|
0
|
my $s = $self->{segments} or return wantarray ? () : 0; |
|
|
|
0
|
|
|
|
|
|
|
299
|
0
|
|
|
|
|
0
|
@$s; |
|
300
|
|
|
|
|
|
|
} |
|
301
|
|
|
|
|
|
|
sub score { |
|
302
|
0
|
|
|
0
|
0
|
0
|
my $self = shift; |
|
303
|
0
|
|
|
|
|
0
|
my $d = $self->{score}; |
|
304
|
0
|
0
|
|
|
|
0
|
$self->{score} = shift if @_; |
|
305
|
0
|
|
|
|
|
0
|
$d; |
|
306
|
|
|
|
|
|
|
} |
|
307
|
|
|
|
|
|
|
sub primary_tag { |
|
308
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
|
309
|
0
|
|
|
|
|
0
|
my $d = $self->{type}; |
|
310
|
0
|
0
|
|
|
|
0
|
$self->{type} = shift if @_; |
|
311
|
0
|
|
|
|
|
0
|
$d; |
|
312
|
|
|
|
|
|
|
} |
|
313
|
|
|
|
|
|
|
sub name { |
|
314
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
|
315
|
0
|
|
|
|
|
0
|
my $d = $self->{name}; |
|
316
|
0
|
0
|
|
|
|
0
|
$self->{name} = shift if @_; |
|
317
|
0
|
|
|
|
|
0
|
$d; |
|
318
|
|
|
|
|
|
|
} |
|
319
|
0
|
|
|
0
|
1
|
0
|
sub seq_id { shift->ref(@_) } |
|
320
|
|
|
|
|
|
|
sub ref { |
|
321
|
0
|
|
|
0
|
0
|
0
|
my $self = shift; |
|
322
|
0
|
|
|
|
|
0
|
my $d = $self->{ref}; |
|
323
|
0
|
0
|
|
|
|
0
|
$self->{ref} = shift if @_; |
|
324
|
0
|
|
|
|
|
0
|
$d; |
|
325
|
|
|
|
|
|
|
} |
|
326
|
|
|
|
|
|
|
sub start { |
|
327
|
0
|
|
|
0
|
1
|
0
|
my $self = shift; |
|
328
|
0
|
|
|
|
|
0
|
my $d = $self->{start}; |
|
329
|
0
|
0
|
|
|
|
0
|
$self->{start} = shift if @_; |
|
330
|
0
|
0
|
|
|
|
0
|
if (my $rs = $self->{refseq}) { |
|
331
|
0
|
|
0
|
|
|
0
|
my $strand = $rs->strand || 1; |
|
332
|
0
|
0
|
|
|
|
0
|
return $strand >= 0 ? ($d - $rs->start + 1) : ($rs->end - $d + 1); |
|
333
|
|
|
|
|
|
|
} else { |
|
334
|
0
|
|
|
|
|
0
|
return $d; |
|
335
|
|
|
|
|
|
|
} |
|
336
|
|
|
|
|
|
|
} |
|
337
|
|
|
|
|
|
|
sub end { |
|
338
|
2
|
|
|
2
|
1
|
3
|
my $self = shift; |
|
339
|
2
|
|
|
|
|
2
|
my $d = $self->{stop}; |
|
340
|
2
|
50
|
|
|
|
6
|
$self->{stop} = shift if @_; |
|
341
|
2
|
50
|
|
|
|
3
|
if (my $rs = $self->{refseq}) { |
|
342
|
0
|
|
0
|
|
|
0
|
my $strand = $rs->strand || 1; |
|
343
|
0
|
0
|
|
|
|
0
|
return $strand >= 0 ? ($d - $rs->start + 1) : ($rs->end - $d + 1); |
|
344
|
|
|
|
|
|
|
} |
|
345
|
2
|
|
|
|
|
6
|
$d; |
|
346
|
|
|
|
|
|
|
} |
|
347
|
|
|
|
|
|
|
sub strand { |
|
348
|
3
|
|
|
3
|
1
|
7
|
my $self = shift; |
|
349
|
3
|
|
|
|
|
3
|
my $d = $self->{strand}; |
|
350
|
3
|
50
|
|
|
|
7
|
$self->{strand} = shift if @_; |
|
351
|
3
|
50
|
|
|
|
6
|
if (my $rs = $self->{refseq}) { |
|
352
|
0
|
|
|
|
|
0
|
my $rstrand = $rs->strand; |
|
353
|
0
|
0
|
|
|
|
0
|
return 0 unless $d; |
|
354
|
0
|
0
|
|
|
|
0
|
return 1 if $rstrand == $d; |
|
355
|
0
|
0
|
|
|
|
0
|
return -1 if $rstrand != $d; |
|
356
|
|
|
|
|
|
|
} |
|
357
|
3
|
|
|
|
|
7
|
$d; |
|
358
|
|
|
|
|
|
|
} |
|
359
|
|
|
|
|
|
|
|
|
360
|
|
|
|
|
|
|
# this does nothing, but it is here for compatibility reasons |
|
361
|
|
|
|
|
|
|
sub absolute { |
|
362
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
363
|
0
|
|
|
|
|
|
my $d = $self->{absolute}; |
|
364
|
0
|
0
|
|
|
|
|
$self->{absolute} = shift if @_; |
|
365
|
0
|
|
|
|
|
|
$d; |
|
366
|
|
|
|
|
|
|
} |
|
367
|
|
|
|
|
|
|
|
|
368
|
|
|
|
|
|
|
sub abs_start { |
|
369
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
370
|
0
|
|
|
|
|
|
local $self->{refseq} = undef; |
|
371
|
0
|
|
|
|
|
|
$self->start(@_); |
|
372
|
|
|
|
|
|
|
} |
|
373
|
|
|
|
|
|
|
sub abs_end { |
|
374
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
375
|
0
|
|
|
|
|
|
local $self->{refseq} = undef; |
|
376
|
0
|
|
|
|
|
|
$self->end(@_); |
|
377
|
|
|
|
|
|
|
} |
|
378
|
|
|
|
|
|
|
sub abs_strand { |
|
379
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
380
|
0
|
|
|
|
|
|
local $self->{refseq} = undef; |
|
381
|
0
|
|
|
|
|
|
$self->strand(@_); |
|
382
|
|
|
|
|
|
|
} |
|
383
|
|
|
|
|
|
|
|
|
384
|
|
|
|
|
|
|
sub length { |
|
385
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
386
|
0
|
|
|
|
|
|
return $self->end - $self->start + 1; |
|
387
|
|
|
|
|
|
|
} |
|
388
|
|
|
|
|
|
|
|
|
389
|
|
|
|
|
|
|
#is_circular is needed for Bio::PrimarySeqI |
|
390
|
|
|
|
|
|
|
sub is_circular { |
|
391
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
392
|
0
|
|
|
|
|
|
my $d = $self->{is_circular}; |
|
393
|
0
|
0
|
|
|
|
|
$self->{is_circular} = shift if @_; |
|
394
|
0
|
|
|
|
|
|
$d; |
|
395
|
|
|
|
|
|
|
} |
|
396
|
|
|
|
|
|
|
|
|
397
|
|
|
|
|
|
|
|
|
398
|
|
|
|
|
|
|
sub seq { |
|
399
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
400
|
0
|
0
|
|
|
|
|
my $seq = exists $self->{seq} ? $self->{seq} : ''; |
|
401
|
0
|
|
|
|
|
|
return $seq; |
|
402
|
|
|
|
|
|
|
} |
|
403
|
|
|
|
|
|
|
|
|
404
|
|
|
|
|
|
|
sub dna { |
|
405
|
0
|
|
|
0
|
0
|
|
my $seq = shift->seq; |
|
406
|
0
|
0
|
|
|
|
|
$seq = $seq->seq if CORE::ref($seq); |
|
407
|
0
|
|
|
|
|
|
return $seq; |
|
408
|
|
|
|
|
|
|
} |
|
409
|
|
|
|
|
|
|
|
|
410
|
|
|
|
|
|
|
=head2 display_name |
|
411
|
|
|
|
|
|
|
|
|
412
|
|
|
|
|
|
|
Title : display_name |
|
413
|
|
|
|
|
|
|
Usage : $id = $obj->display_name or $obj->display_name($newid); |
|
414
|
|
|
|
|
|
|
Function: Gets or sets the display id, also known as the common name of |
|
415
|
|
|
|
|
|
|
the Seq object. |
|
416
|
|
|
|
|
|
|
|
|
417
|
|
|
|
|
|
|
The semantics of this is that it is the most likely string |
|
418
|
|
|
|
|
|
|
to be used as an identifier of the sequence, and likely to |
|
419
|
|
|
|
|
|
|
have "human" readability. The id is equivalent to the LOCUS |
|
420
|
|
|
|
|
|
|
field of the GenBank/EMBL databanks and the ID field of the |
|
421
|
|
|
|
|
|
|
Swissprot/sptrembl database. In fasta format, the >(\S+) is |
|
422
|
|
|
|
|
|
|
presumed to be the id, though some people overload the id |
|
423
|
|
|
|
|
|
|
to embed other information. Bioperl does not use any |
|
424
|
|
|
|
|
|
|
embedded information in the ID field, and people are |
|
425
|
|
|
|
|
|
|
encouraged to use other mechanisms (accession field for |
|
426
|
|
|
|
|
|
|
example, or extending the sequence object) to solve this. |
|
427
|
|
|
|
|
|
|
|
|
428
|
|
|
|
|
|
|
Notice that $seq->id() maps to this function, mainly for |
|
429
|
|
|
|
|
|
|
legacy/convenience issues. |
|
430
|
|
|
|
|
|
|
Returns : A string |
|
431
|
|
|
|
|
|
|
Args : None or a new id |
|
432
|
|
|
|
|
|
|
|
|
433
|
|
|
|
|
|
|
|
|
434
|
|
|
|
|
|
|
=cut |
|
435
|
|
|
|
|
|
|
|
|
436
|
0
|
|
|
0
|
1
|
|
sub display_name { shift->name(@_) } |
|
437
|
|
|
|
|
|
|
|
|
438
|
|
|
|
|
|
|
*display_id = \&display_name; |
|
439
|
|
|
|
|
|
|
|
|
440
|
|
|
|
|
|
|
=head2 accession_number |
|
441
|
|
|
|
|
|
|
|
|
442
|
|
|
|
|
|
|
Title : accession_number |
|
443
|
|
|
|
|
|
|
Usage : $unique_biological_key = $obj->accession_number; |
|
444
|
|
|
|
|
|
|
Function: Returns the unique biological id for a sequence, commonly |
|
445
|
|
|
|
|
|
|
called the accession_number. For sequences from established |
|
446
|
|
|
|
|
|
|
databases, the implementors should try to use the correct |
|
447
|
|
|
|
|
|
|
accession number. Notice that primary_id() provides the |
|
448
|
|
|
|
|
|
|
unique id for the implemetation, allowing multiple objects |
|
449
|
|
|
|
|
|
|
to have the same accession number in a particular implementation. |
|
450
|
|
|
|
|
|
|
|
|
451
|
|
|
|
|
|
|
For sequences with no accession number, this method should return |
|
452
|
|
|
|
|
|
|
"unknown". |
|
453
|
|
|
|
|
|
|
Returns : A string |
|
454
|
|
|
|
|
|
|
Args : None |
|
455
|
|
|
|
|
|
|
|
|
456
|
|
|
|
|
|
|
|
|
457
|
|
|
|
|
|
|
=cut |
|
458
|
|
|
|
|
|
|
|
|
459
|
|
|
|
|
|
|
sub accession_number { |
|
460
|
0
|
|
|
0
|
1
|
|
return 'unknown'; |
|
461
|
|
|
|
|
|
|
} |
|
462
|
|
|
|
|
|
|
|
|
463
|
|
|
|
|
|
|
=head2 alphabet |
|
464
|
|
|
|
|
|
|
|
|
465
|
|
|
|
|
|
|
Title : alphabet |
|
466
|
|
|
|
|
|
|
Usage : if( $obj->alphabet eq 'dna' ) { /Do Something/ } |
|
467
|
|
|
|
|
|
|
Function: Returns the type of sequence being one of |
|
468
|
|
|
|
|
|
|
'dna', 'rna' or 'protein'. This is case sensitive. |
|
469
|
|
|
|
|
|
|
|
|
470
|
|
|
|
|
|
|
This is not called because this would cause |
|
471
|
|
|
|
|
|
|
upgrade problems from the 0.5 and earlier Seq objects. |
|
472
|
|
|
|
|
|
|
|
|
473
|
|
|
|
|
|
|
Returns : a string either 'dna','rna','protein'. NB - the object must |
|
474
|
|
|
|
|
|
|
make a call of the type - if there is no type specified it |
|
475
|
|
|
|
|
|
|
has to guess. |
|
476
|
|
|
|
|
|
|
Args : none |
|
477
|
|
|
|
|
|
|
Status : Virtual |
|
478
|
|
|
|
|
|
|
|
|
479
|
|
|
|
|
|
|
|
|
480
|
|
|
|
|
|
|
=cut |
|
481
|
|
|
|
|
|
|
|
|
482
|
|
|
|
|
|
|
sub alphabet{ |
|
483
|
0
|
|
|
0
|
1
|
|
return 'dna'; # no way this will be anything other than dna! |
|
484
|
|
|
|
|
|
|
} |
|
485
|
|
|
|
|
|
|
|
|
486
|
|
|
|
|
|
|
|
|
487
|
|
|
|
|
|
|
|
|
488
|
|
|
|
|
|
|
=head2 desc |
|
489
|
|
|
|
|
|
|
|
|
490
|
|
|
|
|
|
|
Title : desc |
|
491
|
|
|
|
|
|
|
Usage : $seqobj->desc($string) or $seqobj->desc() |
|
492
|
|
|
|
|
|
|
Function: Sets or gets the description of the sequence |
|
493
|
|
|
|
|
|
|
Example : |
|
494
|
|
|
|
|
|
|
Returns : The description |
|
495
|
|
|
|
|
|
|
Args : The description or none |
|
496
|
|
|
|
|
|
|
|
|
497
|
|
|
|
|
|
|
|
|
498
|
|
|
|
|
|
|
=cut |
|
499
|
|
|
|
|
|
|
|
|
500
|
|
|
|
|
|
|
sub desc { |
|
501
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
502
|
0
|
|
|
|
|
|
my ($d) = $self->notes; |
|
503
|
0
|
0
|
|
|
|
|
$self->{desc} = shift if @_; |
|
504
|
0
|
|
|
|
|
|
$d; |
|
505
|
|
|
|
|
|
|
} |
|
506
|
|
|
|
|
|
|
|
|
507
|
|
|
|
|
|
|
sub attributes { |
|
508
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
509
|
0
|
0
|
|
|
|
|
if (@_) { |
|
510
|
0
|
|
|
|
|
|
return $self->get_tag_values(@_); |
|
511
|
|
|
|
|
|
|
} else { |
|
512
|
0
|
0
|
|
|
|
|
return $self->{attributes} ? %{$self->{attributes}} : (); |
|
|
0
|
|
|
|
|
|
|
|
513
|
|
|
|
|
|
|
} |
|
514
|
|
|
|
|
|
|
} |
|
515
|
|
|
|
|
|
|
|
|
516
|
|
|
|
|
|
|
sub primary_id { |
|
517
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
518
|
0
|
|
|
|
|
|
my $d = $self->{primary_id}; |
|
519
|
0
|
0
|
|
|
|
|
$self->{primary_id} = shift if @_; |
|
520
|
0
|
|
|
|
|
|
return $d; |
|
521
|
|
|
|
|
|
|
# return $d if defined $d; |
|
522
|
|
|
|
|
|
|
# return (overload::StrVal($self) =~ /0x([a-f0-9]+)/)[0]; |
|
523
|
|
|
|
|
|
|
} |
|
524
|
|
|
|
|
|
|
|
|
525
|
|
|
|
|
|
|
sub notes { |
|
526
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
527
|
0
|
|
|
|
|
|
my $notes = $self->{desc}; |
|
528
|
0
|
0
|
|
|
|
|
return $notes if defined $notes; |
|
529
|
0
|
|
|
|
|
|
return $self->attributes('Note'); |
|
530
|
|
|
|
|
|
|
} |
|
531
|
|
|
|
|
|
|
|
|
532
|
|
|
|
|
|
|
sub aliases { |
|
533
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
534
|
0
|
|
|
|
|
|
return $self->attributes('Alias'); |
|
535
|
|
|
|
|
|
|
} |
|
536
|
|
|
|
|
|
|
|
|
537
|
|
|
|
|
|
|
sub low { |
|
538
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
539
|
0
|
0
|
|
|
|
|
return $self->start < $self->end ? $self->start : $self->end; |
|
540
|
|
|
|
|
|
|
} |
|
541
|
|
|
|
|
|
|
|
|
542
|
|
|
|
|
|
|
sub high { |
|
543
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
544
|
0
|
0
|
|
|
|
|
return $self->start > $self->end ? $self->start : $self->end; |
|
545
|
|
|
|
|
|
|
} |
|
546
|
|
|
|
|
|
|
|
|
547
|
|
|
|
|
|
|
=head2 location |
|
548
|
|
|
|
|
|
|
|
|
549
|
|
|
|
|
|
|
Title : location |
|
550
|
|
|
|
|
|
|
Usage : my $location = $seqfeature->location() |
|
551
|
|
|
|
|
|
|
Function: returns a location object suitable for identifying location |
|
552
|
|
|
|
|
|
|
of feature on sequence or parent feature |
|
553
|
|
|
|
|
|
|
Returns : Bio::LocationI object |
|
554
|
|
|
|
|
|
|
Args : none |
|
555
|
|
|
|
|
|
|
|
|
556
|
|
|
|
|
|
|
=cut |
|
557
|
|
|
|
|
|
|
|
|
558
|
|
|
|
|
|
|
sub location { |
|
559
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
560
|
0
|
0
|
|
|
|
|
require Bio::Location::Split unless Bio::Location::Split->can('new'); |
|
561
|
0
|
|
|
|
|
|
my $location; |
|
562
|
0
|
0
|
|
|
|
|
if (my @segments = $self->segments) { |
|
563
|
0
|
|
|
|
|
|
$location = Bio::Location::Split->new(); |
|
564
|
0
|
|
|
|
|
|
foreach (@segments) { |
|
565
|
0
|
|
|
|
|
|
$location->add_sub_Location($_); |
|
566
|
|
|
|
|
|
|
} |
|
567
|
|
|
|
|
|
|
} else { |
|
568
|
0
|
|
|
|
|
|
$location = $self; |
|
569
|
|
|
|
|
|
|
} |
|
570
|
0
|
|
|
|
|
|
$location; |
|
571
|
|
|
|
|
|
|
} |
|
572
|
|
|
|
|
|
|
|
|
573
|
|
|
|
|
|
|
sub each_Location { |
|
574
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
575
|
0
|
0
|
|
|
|
|
require Bio::Location::Simple unless Bio::Location::Simple->can('new'); |
|
576
|
0
|
0
|
|
|
|
|
if (my @segments = $self->segments) { |
|
577
|
|
|
|
|
|
|
return map { |
|
578
|
0
|
|
|
|
|
|
Bio::Location::Simple->new(-start => $_->start, |
|
|
0
|
|
|
|
|
|
|
|
579
|
|
|
|
|
|
|
-end => $_->end, |
|
580
|
|
|
|
|
|
|
-strand => $_->strand); |
|
581
|
|
|
|
|
|
|
} @segments; |
|
582
|
|
|
|
|
|
|
} else { |
|
583
|
0
|
|
|
|
|
|
return Bio::Location::Simple->new(-start => $self->start, |
|
584
|
|
|
|
|
|
|
-end => $self->end, |
|
585
|
|
|
|
|
|
|
-strand => $self->strand); |
|
586
|
|
|
|
|
|
|
} |
|
587
|
|
|
|
|
|
|
} |
|
588
|
|
|
|
|
|
|
|
|
589
|
|
|
|
|
|
|
=head2 location_string |
|
590
|
|
|
|
|
|
|
|
|
591
|
|
|
|
|
|
|
Title : location_string |
|
592
|
|
|
|
|
|
|
Usage : my $string = $seqfeature->location_string() |
|
593
|
|
|
|
|
|
|
Function: Returns a location string in a format recognized by gbrowse |
|
594
|
|
|
|
|
|
|
Returns : a string |
|
595
|
|
|
|
|
|
|
Args : none |
|
596
|
|
|
|
|
|
|
|
|
597
|
|
|
|
|
|
|
This is a convenience function used by the generic genome browser. It |
|
598
|
|
|
|
|
|
|
returns the location of the feature and its subfeatures in the compact |
|
599
|
|
|
|
|
|
|
form "start1..end1,start2..end2,...". Use |
|
600
|
|
|
|
|
|
|
$seqfeature-Elocation()-EtoFTString() to obtain a standard |
|
601
|
|
|
|
|
|
|
GenBank/EMBL location representation. |
|
602
|
|
|
|
|
|
|
|
|
603
|
|
|
|
|
|
|
=cut |
|
604
|
|
|
|
|
|
|
|
|
605
|
|
|
|
|
|
|
sub location_string { |
|
606
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
607
|
0
|
0
|
|
|
|
|
my @segments = $self->segments or return $self->to_FTstring; |
|
608
|
0
|
|
|
|
|
|
join ',',map {$_->to_FTstring} @segments; |
|
|
0
|
|
|
|
|
|
|
|
609
|
|
|
|
|
|
|
} |
|
610
|
|
|
|
|
|
|
|
|
611
|
|
|
|
|
|
|
sub coordinate_policy { |
|
612
|
0
|
0
|
|
0
|
1
|
|
require Bio::Location::WidestCoordPolicy unless Bio::Location::WidestCoordPolicy->can('new'); |
|
613
|
0
|
|
|
|
|
|
return Bio::Location::WidestCoordPolicy->new(); |
|
614
|
|
|
|
|
|
|
} |
|
615
|
|
|
|
|
|
|
|
|
616
|
0
|
|
|
0
|
1
|
|
sub min_start { shift->low } |
|
617
|
0
|
|
|
0
|
1
|
|
sub max_start { shift->low } |
|
618
|
0
|
|
|
0
|
1
|
|
sub min_end { shift->high } |
|
619
|
0
|
|
|
0
|
1
|
|
sub max_end { shift->high} |
|
620
|
0
|
|
|
0
|
1
|
|
sub start_pos_type { 'EXACT' } |
|
621
|
0
|
|
|
0
|
1
|
|
sub end_pos_type { 'EXACT' } |
|
622
|
|
|
|
|
|
|
sub to_FTstring { |
|
623
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
624
|
0
|
|
|
|
|
|
my $low = $self->min_start; |
|
625
|
0
|
|
|
|
|
|
my $high = $self->max_end; |
|
626
|
0
|
|
|
|
|
|
my $strand = $self->strand; |
|
627
|
0
|
0
|
0
|
|
|
|
my $str = defined $strand && $strand<0 ? "complement($low..$high)" : "$low..$high"; |
|
628
|
0
|
0
|
|
|
|
|
if (my $id = $self->seq_id()) { |
|
629
|
0
|
|
|
|
|
|
$str = $id . ":" . $str; |
|
630
|
|
|
|
|
|
|
} |
|
631
|
0
|
|
|
|
|
|
$str; |
|
632
|
|
|
|
|
|
|
} |
|
633
|
|
|
|
|
|
|
sub phase { |
|
634
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
635
|
0
|
|
|
|
|
|
my $d = $self->{phase}; |
|
636
|
0
|
0
|
|
|
|
|
$self->{phase} = shift if @_; |
|
637
|
0
|
|
|
|
|
|
$d; |
|
638
|
|
|
|
|
|
|
} |
|
639
|
|
|
|
|
|
|
|
|
640
|
|
|
|
|
|
|
sub class { |
|
641
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
642
|
0
|
|
|
|
|
|
my $d = $self->{class}; |
|
643
|
0
|
0
|
|
|
|
|
$self->{class} = shift if @_; |
|
644
|
0
|
0
|
|
|
|
|
return defined($d) ? $d : 'Sequence'; # acedb is still haunting me - LS |
|
645
|
|
|
|
|
|
|
} |
|
646
|
|
|
|
|
|
|
|
|
647
|
|
|
|
|
|
|
# set GFF dumping version |
|
648
|
|
|
|
|
|
|
sub version { |
|
649
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
650
|
0
|
|
0
|
|
|
|
my $d = $self->{gff3_version} || 2; |
|
651
|
0
|
0
|
|
|
|
|
$self->{gff3_version} = shift if @_; |
|
652
|
0
|
|
|
|
|
|
$d; |
|
653
|
|
|
|
|
|
|
} |
|
654
|
|
|
|
|
|
|
|
|
655
|
|
|
|
|
|
|
sub gff_string { |
|
656
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
657
|
|
|
|
|
|
|
|
|
658
|
0
|
0
|
|
|
|
|
if ($self->version == 3) { |
|
659
|
0
|
|
|
|
|
|
return $self->gff3_string(@_); |
|
660
|
|
|
|
|
|
|
} |
|
661
|
|
|
|
|
|
|
|
|
662
|
0
|
|
|
|
|
|
my $recurse = shift; |
|
663
|
0
|
|
|
|
|
|
my $name = $self->name; |
|
664
|
0
|
|
|
|
|
|
my $class = $self->class; |
|
665
|
0
|
0
|
|
|
|
|
my $group = "$class $name" if $name; |
|
666
|
0
|
|
|
|
|
|
my $strand = ('-','.','+')[$self->strand+1]; |
|
667
|
0
|
|
|
|
|
|
my $string; |
|
668
|
0
|
0
|
0
|
|
|
|
$string .= join("\t", |
|
|
|
0
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
669
|
|
|
|
|
|
|
$self->ref||'.',$self->source||'.',$self->method||'.', |
|
670
|
|
|
|
|
|
|
$self->start||'.',$self->stop||'.', |
|
671
|
|
|
|
|
|
|
defined($self->score) ? $self->score : '.', |
|
672
|
|
|
|
|
|
|
$strand||'.', |
|
673
|
|
|
|
|
|
|
defined($self->phase) ? $self->phase : '.', |
|
674
|
|
|
|
|
|
|
$group||'' |
|
675
|
|
|
|
|
|
|
); |
|
676
|
0
|
|
|
|
|
|
$string .= "\n"; |
|
677
|
0
|
0
|
|
|
|
|
if ($recurse) { |
|
678
|
0
|
|
|
|
|
|
foreach ($self->sub_SeqFeature) { |
|
679
|
0
|
|
|
|
|
|
$string .= $_->gff_string($recurse); |
|
680
|
|
|
|
|
|
|
} |
|
681
|
|
|
|
|
|
|
} |
|
682
|
0
|
|
|
|
|
|
$string; |
|
683
|
|
|
|
|
|
|
} |
|
684
|
|
|
|
|
|
|
|
|
685
|
|
|
|
|
|
|
# Suggested strategy for dealing with the multiple parentage issue. |
|
686
|
|
|
|
|
|
|
# First recurse through object tree and record parent tree. |
|
687
|
|
|
|
|
|
|
# Then recurse again, skipping objects we've seen before. |
|
688
|
|
|
|
|
|
|
sub gff3_string { |
|
689
|
0
|
|
|
0
|
0
|
|
my ($self,$recurse,$parent_tree,$seenit,$force_id) = @_; |
|
690
|
0
|
|
0
|
|
|
|
$parent_tree ||= {}; |
|
691
|
0
|
|
0
|
|
|
|
$seenit ||= {}; |
|
692
|
0
|
|
|
|
|
|
my @rsf = (); |
|
693
|
0
|
|
|
|
|
|
my @parent_ids; |
|
694
|
|
|
|
|
|
|
|
|
695
|
0
|
0
|
|
|
|
|
if ($recurse) { |
|
696
|
0
|
0
|
|
|
|
|
$self->_traverse($parent_tree) unless %$parent_tree; # this will record parents of all children |
|
697
|
0
|
0
|
|
|
|
|
my $primary_id = defined $force_id ? $force_id : $self->_real_or_dummy_id; |
|
698
|
|
|
|
|
|
|
|
|
699
|
0
|
0
|
|
|
|
|
return if $seenit->{$primary_id}++; |
|
700
|
|
|
|
|
|
|
|
|
701
|
0
|
|
|
|
|
|
@rsf = $self->get_SeqFeatures; |
|
702
|
0
|
0
|
|
|
|
|
if (@rsf) { |
|
703
|
|
|
|
|
|
|
# Detect case in which we have a split location feature. In this case we |
|
704
|
|
|
|
|
|
|
# skip to the grandchildren and trick them into thinking that our parent is theirs. |
|
705
|
0
|
|
|
|
|
|
my %types = map {$_->primary_tag=>1} @rsf; |
|
|
0
|
|
|
|
|
|
|
|
706
|
0
|
|
|
|
|
|
my @types = keys %types; |
|
707
|
0
|
0
|
0
|
|
|
|
if (@types == 1 && $types[0] eq $self->primary_tag) { |
|
708
|
0
|
|
|
|
|
|
return join ("\n",map {$_->gff3_string(1,$parent_tree,{},$primary_id)} @rsf); |
|
|
0
|
|
|
|
|
|
|
|
709
|
|
|
|
|
|
|
} |
|
710
|
|
|
|
|
|
|
} |
|
711
|
|
|
|
|
|
|
|
|
712
|
0
|
|
|
|
|
|
@parent_ids = keys %{$parent_tree->{$primary_id}}; |
|
|
0
|
|
|
|
|
|
|
|
713
|
|
|
|
|
|
|
} |
|
714
|
|
|
|
|
|
|
|
|
715
|
0
|
|
|
|
|
|
my $group = $self->format_attributes(\@parent_ids,$force_id); |
|
716
|
0
|
|
|
|
|
|
my $name = $self->name; |
|
717
|
|
|
|
|
|
|
|
|
718
|
0
|
|
|
|
|
|
my $class = $self->class; |
|
719
|
0
|
|
|
|
|
|
my $strand = ('-','.','+')[$self->strand+1]; |
|
720
|
0
|
0
|
0
|
|
|
|
my $p = join("\t", |
|
|
|
0
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
721
|
|
|
|
|
|
|
$self->seq_id||'.', |
|
722
|
|
|
|
|
|
|
$self->source||'.', |
|
723
|
|
|
|
|
|
|
$self->method||'.', |
|
724
|
|
|
|
|
|
|
$self->start||'.', |
|
725
|
|
|
|
|
|
|
$self->stop||'.', |
|
726
|
|
|
|
|
|
|
defined($self->score) ? $self->score : '.', |
|
727
|
|
|
|
|
|
|
$strand||'.', |
|
728
|
|
|
|
|
|
|
defined($self->phase) ? $self->phase : '.', |
|
729
|
|
|
|
|
|
|
$group||''); |
|
730
|
|
|
|
|
|
|
return join("\n", |
|
731
|
|
|
|
|
|
|
$p, |
|
732
|
0
|
|
|
|
|
|
map {$_->gff3_string(1,$parent_tree,$seenit)} @rsf); |
|
|
0
|
|
|
|
|
|
|
|
733
|
|
|
|
|
|
|
} |
|
734
|
|
|
|
|
|
|
|
|
735
|
|
|
|
|
|
|
sub _real_or_dummy_id { |
|
736
|
0
|
|
|
0
|
|
|
my $self = shift; |
|
737
|
0
|
|
|
|
|
|
my $id = $self->primary_id; |
|
738
|
0
|
0
|
|
|
|
|
return $id if defined $id; |
|
739
|
0
|
|
|
|
|
|
return return (overload::StrVal($self) =~ /0x([a-f0-9]+)/)[0]; |
|
740
|
|
|
|
|
|
|
} |
|
741
|
|
|
|
|
|
|
|
|
742
|
|
|
|
|
|
|
sub _traverse { |
|
743
|
0
|
|
|
0
|
|
|
my $self = shift; |
|
744
|
0
|
|
|
|
|
|
my $tree = shift; # tree => {$child}{$parent} = 1 |
|
745
|
0
|
|
|
|
|
|
my $parent = shift; |
|
746
|
0
|
|
|
|
|
|
my $id = $self->_real_or_dummy_id; |
|
747
|
0
|
0
|
|
|
|
|
defined $id or return; |
|
748
|
0
|
0
|
|
|
|
|
$tree->{$id}{$parent->_real_or_dummy_id}++ if $parent; |
|
749
|
0
|
|
|
|
|
|
$_->_traverse($tree,$self) foreach $self->get_SeqFeatures; |
|
750
|
|
|
|
|
|
|
} |
|
751
|
|
|
|
|
|
|
|
|
752
|
0
|
|
|
0
|
0
|
|
sub db { return } |
|
753
|
|
|
|
|
|
|
|
|
754
|
|
|
|
|
|
|
sub source_tag { |
|
755
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
756
|
0
|
|
|
|
|
|
my $d = $self->{source}; |
|
757
|
0
|
0
|
|
|
|
|
$self->{source} = shift if @_; |
|
758
|
0
|
|
|
|
|
|
$d; |
|
759
|
|
|
|
|
|
|
} |
|
760
|
|
|
|
|
|
|
|
|
761
|
|
|
|
|
|
|
# This probably should be deleted. Not sure why it's here, but might |
|
762
|
|
|
|
|
|
|
# have been added for Ace::Sequence::Feature-compliance. |
|
763
|
|
|
|
|
|
|
sub introns { |
|
764
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
765
|
0
|
|
|
|
|
|
return; |
|
766
|
|
|
|
|
|
|
} |
|
767
|
|
|
|
|
|
|
|
|
768
|
|
|
|
|
|
|
sub has_tag { |
|
769
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
770
|
0
|
|
|
|
|
|
my $tag = shift; |
|
771
|
0
|
|
|
|
|
|
return exists $self->{attributes}{$tag}; |
|
772
|
|
|
|
|
|
|
} |
|
773
|
|
|
|
|
|
|
|
|
774
|
|
|
|
|
|
|
sub escape { |
|
775
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
776
|
0
|
|
|
|
|
|
my $toencode = shift; |
|
777
|
0
|
|
|
|
|
|
$toencode =~ s/([^a-zA-Z0-9_.:?^*\(\)\[\]@!+-])/uc sprintf("%%%02x",ord($1))/eg; |
|
|
0
|
|
|
|
|
|
|
|
778
|
0
|
|
|
|
|
|
$toencode; |
|
779
|
|
|
|
|
|
|
} |
|
780
|
|
|
|
|
|
|
|
|
781
|
|
|
|
|
|
|
sub all_tags { |
|
782
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
783
|
0
|
|
|
|
|
|
return keys %{$self->{attributes}}; |
|
|
0
|
|
|
|
|
|
|
|
784
|
|
|
|
|
|
|
} |
|
785
|
|
|
|
|
|
|
|
|
786
|
|
|
|
|
|
|
sub add_tag_value { |
|
787
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
788
|
0
|
|
|
|
|
|
my ($tag_name,@tag_values) = @_; |
|
789
|
0
|
|
|
|
|
|
push @{$self->{attributes}{$tag_name}},@tag_values; |
|
|
0
|
|
|
|
|
|
|
|
790
|
|
|
|
|
|
|
} |
|
791
|
|
|
|
|
|
|
|
|
792
|
|
|
|
|
|
|
sub remove_tag { |
|
793
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
794
|
0
|
|
|
|
|
|
my $tag_name = shift; |
|
795
|
0
|
|
|
|
|
|
delete $self->{attributes}{$tag_name}; |
|
796
|
|
|
|
|
|
|
} |
|
797
|
|
|
|
|
|
|
|
|
798
|
|
|
|
|
|
|
sub each_tag_value { |
|
799
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
800
|
0
|
|
|
|
|
|
my $tag = shift; |
|
801
|
0
|
0
|
|
|
|
|
my $value = $self->{attributes}{$tag} or return; |
|
802
|
0
|
|
|
|
|
|
my $ref = CORE::ref $value; |
|
803
|
0
|
|
|
|
|
|
return $ref && $ref eq 'ARRAY' ? @{$self->{attributes}{$tag}} |
|
804
|
0
|
0
|
0
|
|
|
|
: $self->{attributes}{$tag}; |
|
805
|
|
|
|
|
|
|
} |
|
806
|
|
|
|
|
|
|
|
|
807
|
|
|
|
|
|
|
sub get_Annotations { |
|
808
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
809
|
0
|
|
|
|
|
|
my $tag = shift; |
|
810
|
0
|
|
|
|
|
|
my @values = $self->get_tag_values($tag); |
|
811
|
0
|
0
|
|
|
|
|
return $values[0] if @values == 1; |
|
812
|
0
|
|
|
|
|
|
return @values; |
|
813
|
|
|
|
|
|
|
} |
|
814
|
|
|
|
|
|
|
|
|
815
|
|
|
|
|
|
|
sub format_attributes { |
|
816
|
0
|
|
|
0
|
0
|
|
my $self = shift; |
|
817
|
0
|
|
|
|
|
|
my $parent = shift; |
|
818
|
0
|
|
|
|
|
|
my $fallback_id = shift; |
|
819
|
|
|
|
|
|
|
|
|
820
|
0
|
|
|
|
|
|
my @tags = $self->get_all_tags; |
|
821
|
0
|
|
|
|
|
|
my @result; |
|
822
|
0
|
|
|
|
|
|
for my $t (@tags) { |
|
823
|
0
|
|
|
|
|
|
my @values = $self->get_tag_values($t); |
|
824
|
0
|
0
|
|
|
|
|
push @result,join '=',$self->escape($t),join(',', map {$self->escape($_)} @values) if @values; |
|
|
0
|
|
|
|
|
|
|
|
825
|
|
|
|
|
|
|
} |
|
826
|
|
|
|
|
|
|
#my $id = $self->escape($self->_real_or_dummy_id) || $fallback_id; |
|
827
|
0
|
|
0
|
|
|
|
my $id = $fallback_id || $self->escape($self->_real_or_dummy_id); |
|
828
|
|
|
|
|
|
|
|
|
829
|
0
|
|
|
|
|
|
my $parent_id; |
|
830
|
0
|
0
|
|
|
|
|
if (@$parent) { |
|
831
|
0
|
|
|
|
|
|
$parent_id = join (',',map {$self->escape($_)} @$parent); |
|
|
0
|
|
|
|
|
|
|
|
832
|
|
|
|
|
|
|
} |
|
833
|
|
|
|
|
|
|
|
|
834
|
0
|
|
|
|
|
|
my $name = $self->display_name; |
|
835
|
0
|
0
|
|
|
|
|
unshift @result,"ID=".$id if defined $id; |
|
836
|
0
|
0
|
|
|
|
|
unshift @result,"Parent=".$parent_id if defined $parent_id; |
|
837
|
0
|
0
|
|
|
|
|
unshift @result,"Name=".$self->escape($name) if defined $name; |
|
838
|
0
|
|
|
|
|
|
return join ';',@result; |
|
839
|
|
|
|
|
|
|
} |
|
840
|
|
|
|
|
|
|
|
|
841
|
|
|
|
|
|
|
=head2 clone |
|
842
|
|
|
|
|
|
|
|
|
843
|
|
|
|
|
|
|
Title : clone |
|
844
|
|
|
|
|
|
|
Usage : my $feature = $seqfeature->clone |
|
845
|
|
|
|
|
|
|
Function: Create a deep copy of the feature |
|
846
|
|
|
|
|
|
|
Returns : A copy of the feature |
|
847
|
|
|
|
|
|
|
Args : none |
|
848
|
|
|
|
|
|
|
|
|
849
|
|
|
|
|
|
|
=cut |
|
850
|
|
|
|
|
|
|
|
|
851
|
|
|
|
|
|
|
sub clone { |
|
852
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
853
|
0
|
|
|
|
|
|
my %clone = %$self; |
|
854
|
|
|
|
|
|
|
# overwrite attributes |
|
855
|
0
|
|
|
|
|
|
my $clone = bless \%clone,CORE::ref($self); |
|
856
|
0
|
|
|
|
|
|
$clone{attributes} = {}; |
|
857
|
0
|
|
|
|
|
|
for my $k (keys %{$self->{attributes}}) { |
|
|
0
|
|
|
|
|
|
|
|
858
|
0
|
|
|
|
|
|
@{$clone{attributes}{$k}} = @{$self->{attributes}{$k}}; |
|
|
0
|
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
859
|
|
|
|
|
|
|
} |
|
860
|
0
|
|
|
|
|
|
return $clone; |
|
861
|
|
|
|
|
|
|
} |
|
862
|
|
|
|
|
|
|
|
|
863
|
|
|
|
|
|
|
=head2 refseq |
|
864
|
|
|
|
|
|
|
|
|
865
|
|
|
|
|
|
|
Title : refseq |
|
866
|
|
|
|
|
|
|
Usage : $ref = $s->refseq([$newseq] [,$newseqclass]) |
|
867
|
|
|
|
|
|
|
Function: get/set reference sequence |
|
868
|
|
|
|
|
|
|
Returns : current reference sequence |
|
869
|
|
|
|
|
|
|
Args : new reference sequence and class (optional) |
|
870
|
|
|
|
|
|
|
Status : Public |
|
871
|
|
|
|
|
|
|
|
|
872
|
|
|
|
|
|
|
This method will get or set the reference sequence. Called with no |
|
873
|
|
|
|
|
|
|
arguments, it returns the current reference sequence. Called with any |
|
874
|
|
|
|
|
|
|
Bio::SeqFeatureI object that provides the seq_id(), start(), end() and |
|
875
|
|
|
|
|
|
|
strand() methods. |
|
876
|
|
|
|
|
|
|
|
|
877
|
|
|
|
|
|
|
The method will generate an exception if you attempt to set the |
|
878
|
|
|
|
|
|
|
reference sequence to a sequence that has a different seq_id from the |
|
879
|
|
|
|
|
|
|
current feature. |
|
880
|
|
|
|
|
|
|
|
|
881
|
|
|
|
|
|
|
=cut |
|
882
|
|
|
|
|
|
|
|
|
883
|
|
|
|
|
|
|
sub refseq { |
|
884
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
|
885
|
0
|
|
|
|
|
|
my $d = $self->{refseq}; |
|
886
|
0
|
0
|
|
|
|
|
if (@_) { |
|
887
|
0
|
|
|
|
|
|
my $newref = shift; |
|
888
|
0
|
0
|
|
|
|
|
$self->throw("attempt to set refseq using a feature that does not share the same seq_id") |
|
889
|
|
|
|
|
|
|
unless $newref->seq_id eq $self->seq_id; |
|
890
|
0
|
|
|
|
|
|
$self->{refseq} = $newref; |
|
891
|
|
|
|
|
|
|
} |
|
892
|
0
|
|
|
|
|
|
return $d; |
|
893
|
|
|
|
|
|
|
} |
|
894
|
|
|
|
|
|
|
|
|
895
|
|
|
|
0
|
|
|
sub DESTROY { } |
|
896
|
|
|
|
|
|
|
|
|
897
|
|
|
|
|
|
|
1; |
|
898
|
|
|
|
|
|
|
|
|
899
|
|
|
|
|
|
|
__END__ |