line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
=head1 NAME |
2
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
Bio::GeneDesign::CodonJuggle |
4
|
|
|
|
|
|
|
|
5
|
|
|
|
|
|
|
=head1 VERSION |
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
Version 5.56 |
8
|
|
|
|
|
|
|
|
9
|
|
|
|
|
|
|
=head1 DESCRIPTION |
10
|
|
|
|
|
|
|
|
11
|
|
|
|
|
|
|
Codon Juggle a sequence using rscu data to determine replacement likelihood |
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
=head1 AUTHOR |
14
|
|
|
|
|
|
|
|
15
|
|
|
|
|
|
|
Sarah Richardson . |
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
=cut |
18
|
|
|
|
|
|
|
|
19
|
|
|
|
|
|
|
package Bio::GeneDesign::CodonJuggle; |
20
|
|
|
|
|
|
|
require Exporter; |
21
|
|
|
|
|
|
|
|
22
|
11
|
|
|
11
|
|
4083
|
use Bio::GeneDesign::Random qw(_random_index _weighted_rand); |
|
11
|
|
|
|
|
31
|
|
|
11
|
|
|
|
|
631
|
|
23
|
11
|
|
|
11
|
|
65
|
use Bio::GeneDesign::Basic qw(_compare_sequences); |
|
11
|
|
|
|
|
22
|
|
|
11
|
|
|
|
|
411
|
|
24
|
11
|
|
|
11
|
|
56
|
use Carp; |
|
11
|
|
|
|
|
21
|
|
|
11
|
|
|
|
|
443
|
|
25
|
|
|
|
|
|
|
|
26
|
11
|
|
|
11
|
|
56
|
use strict; |
|
11
|
|
|
|
|
18
|
|
|
11
|
|
|
|
|
196
|
|
27
|
11
|
|
|
11
|
|
52
|
use warnings; |
|
11
|
|
|
|
|
16
|
|
|
11
|
|
|
|
|
375
|
|
28
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
our $VERSION = 5.56; |
30
|
|
|
|
|
|
|
|
31
|
11
|
|
|
11
|
|
82
|
use base qw(Exporter); |
|
11
|
|
|
|
|
17
|
|
|
11
|
|
|
|
|
10942
|
|
32
|
|
|
|
|
|
|
our @EXPORT_OK = qw( |
33
|
|
|
|
|
|
|
_list_algorithms |
34
|
|
|
|
|
|
|
_codonJuggle_balanced |
35
|
|
|
|
|
|
|
_codonJuggle_high |
36
|
|
|
|
|
|
|
_codonJuggle_least_different_rscu |
37
|
|
|
|
|
|
|
_codonJuggle_most_different_sequence |
38
|
|
|
|
|
|
|
_codonJuggle_random |
39
|
|
|
|
|
|
|
); |
40
|
|
|
|
|
|
|
our %EXPORT_TAGS = (GD => \@EXPORT_OK); |
41
|
|
|
|
|
|
|
|
42
|
|
|
|
|
|
|
=head2 _list_algorithms |
43
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
=cut |
45
|
|
|
|
|
|
|
|
46
|
|
|
|
|
|
|
sub _list_algorithms |
47
|
|
|
|
|
|
|
{ |
48
|
0
|
|
|
0
|
|
0
|
my %hsh = |
49
|
|
|
|
|
|
|
( |
50
|
|
|
|
|
|
|
balanced => 'weighted averages', |
51
|
|
|
|
|
|
|
high => 'uniformly most used', |
52
|
|
|
|
|
|
|
least_different_RSCU => 'least different by RSCU value', |
53
|
|
|
|
|
|
|
most_different_sequence => 'most different by sequence identity', |
54
|
|
|
|
|
|
|
random => 'random', |
55
|
|
|
|
|
|
|
); |
56
|
0
|
|
|
|
|
0
|
return \%hsh; |
57
|
|
|
|
|
|
|
} |
58
|
|
|
|
|
|
|
|
59
|
|
|
|
|
|
|
=head2 _codonJuggle_balanced |
60
|
|
|
|
|
|
|
|
61
|
|
|
|
|
|
|
=cut |
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
sub _codonJuggle_balanced |
64
|
|
|
|
|
|
|
{ |
65
|
10000
|
|
|
10000
|
|
16285
|
my ($codon_table, $reverse_codon_table, $rscu_table, $nucseq) = @_; |
66
|
10000
|
|
|
|
|
13498
|
$nucseq = uc $nucseq; |
67
|
10000
|
|
|
|
|
13465
|
my %changehsh = (); |
68
|
10000
|
|
|
|
|
33641
|
foreach my $aa (keys %$reverse_codon_table) |
69
|
|
|
|
|
|
|
{ |
70
|
210000
|
|
|
|
|
260747
|
$changehsh{$aa} = {}; |
71
|
210000
|
|
|
|
|
224472
|
my @codons = @{$reverse_codon_table->{$aa}}; |
|
210000
|
|
|
|
|
323636
|
|
72
|
210000
|
|
|
|
|
244973
|
my $count = scalar(@codons); |
73
|
210000
|
|
|
|
|
219862
|
my $checksum = 0; |
74
|
210000
|
|
|
|
|
243428
|
foreach my $coda (@codons) |
75
|
|
|
|
|
|
|
{ |
76
|
640000
|
|
|
|
|
722994
|
my $likely = ($rscu_table->{$coda}) / $count; |
77
|
640000
|
|
|
|
|
774688
|
$changehsh{$aa}->{$coda} = $likely; |
78
|
640000
|
|
|
|
|
749442
|
$checksum += $likely; |
79
|
|
|
|
|
|
|
} |
80
|
210000
|
50
|
|
|
|
353752
|
if ($checksum == 0) |
81
|
|
|
|
|
|
|
{ |
82
|
0
|
|
|
|
|
0
|
croak "This RSCU table has no positive values for $aa\n"; |
83
|
|
|
|
|
|
|
} |
84
|
|
|
|
|
|
|
} |
85
|
10000
|
|
|
|
|
21541
|
my $offset = 0; |
86
|
10000
|
|
|
|
|
12127
|
my $newseq = q{}; |
87
|
10000
|
|
|
|
|
15231
|
while ($offset < length($nucseq)) |
88
|
|
|
|
|
|
|
{ |
89
|
50000
|
|
|
|
|
65459
|
my $curcod = substr($nucseq, $offset, 3); |
90
|
50000
|
|
|
|
|
65295
|
my $aa = $codon_table->{$curcod}; |
91
|
50000
|
|
|
|
|
81649
|
my $newcod = _weighted_rand($changehsh{$aa}); |
92
|
50000
|
|
|
|
|
68727
|
$newseq .= $newcod; |
93
|
50000
|
|
|
|
|
77701
|
$offset += 3; |
94
|
|
|
|
|
|
|
} |
95
|
10000
|
|
|
|
|
70857
|
return $newseq; |
96
|
|
|
|
|
|
|
} |
97
|
|
|
|
|
|
|
|
98
|
|
|
|
|
|
|
=head2 _codonJuggle_high |
99
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
|
101
|
|
|
|
|
|
|
=cut |
102
|
|
|
|
|
|
|
|
103
|
|
|
|
|
|
|
sub _codonJuggle_high |
104
|
|
|
|
|
|
|
{ |
105
|
1
|
|
|
1
|
|
4
|
my ($codon_table, $reverse_codon_table, $rscu_table, $nucseq) = @_; |
106
|
1
|
|
|
|
|
2
|
my $cod_highs = {}; |
107
|
1
|
|
|
|
|
2
|
foreach my $aa (keys %{$reverse_codon_table}) |
|
1
|
|
|
|
|
7
|
|
108
|
|
|
|
|
|
|
{ |
109
|
21
|
|
|
|
|
23
|
my $myrscu = -1; |
110
|
21
|
|
|
|
|
67
|
foreach my $codon (@{$reverse_codon_table->{$aa}}) |
|
21
|
|
|
|
|
32
|
|
111
|
|
|
|
|
|
|
{ |
112
|
64
|
100
|
|
|
|
138
|
if ($rscu_table->{$codon} > $myrscu) |
113
|
|
|
|
|
|
|
{ |
114
|
35
|
|
|
|
|
51
|
$cod_highs->{$aa} = $codon; |
115
|
35
|
|
|
|
|
45
|
$myrscu = $rscu_table->{$codon}; |
116
|
|
|
|
|
|
|
} |
117
|
|
|
|
|
|
|
} |
118
|
|
|
|
|
|
|
} |
119
|
1
|
|
|
|
|
2
|
my $offset = 0; |
120
|
1
|
|
|
|
|
2
|
my $newseq = q{}; |
121
|
1
|
|
|
|
|
3
|
while ($offset < length($nucseq)) |
122
|
|
|
|
|
|
|
{ |
123
|
200
|
|
|
|
|
224
|
my $curcod = substr($nucseq, $offset, 3); |
124
|
200
|
|
|
|
|
248
|
$newseq .= $cod_highs->{$codon_table->{$curcod}}; |
125
|
200
|
|
|
|
|
279
|
$offset += 3; |
126
|
|
|
|
|
|
|
} |
127
|
1
|
|
|
|
|
7
|
return $newseq; |
128
|
|
|
|
|
|
|
} |
129
|
|
|
|
|
|
|
|
130
|
|
|
|
|
|
|
=head2 _codonJuggle_least_different_rscu |
131
|
|
|
|
|
|
|
|
132
|
|
|
|
|
|
|
=cut |
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
sub _codonJuggle_least_different_rscu |
135
|
|
|
|
|
|
|
{ |
136
|
1
|
|
|
1
|
|
3
|
my ($codon_table, $reverse_codon_table, $rscu_table, $nucseq) = @_; |
137
|
1
|
|
|
|
|
4
|
my %changehsh = (); |
138
|
1
|
|
|
|
|
9
|
foreach my $aa (sort keys %$reverse_codon_table) |
139
|
|
|
|
|
|
|
{ |
140
|
21
|
|
|
|
|
28
|
foreach my $coda (@{$reverse_codon_table->{$aa}}) |
|
21
|
|
|
|
|
35
|
|
141
|
|
|
|
|
|
|
{ |
142
|
64
|
|
|
|
|
84
|
$changehsh{$coda} = $coda; |
143
|
|
|
|
|
|
|
my @posarr = sort {abs($rscu_table->{$a} - $rscu_table->{$coda}) |
144
|
69
|
|
|
|
|
131
|
<=> abs($rscu_table->{$b} - $rscu_table->{$coda})} |
145
|
180
|
|
|
|
|
317
|
grep {abs($rscu_table->{$_} - $rscu_table->{$coda}) <= 1} |
146
|
244
|
|
|
|
|
329
|
grep {$_ ne $coda} |
147
|
64
|
|
|
|
|
68
|
@{$reverse_codon_table->{$aa}}; |
|
64
|
|
|
|
|
88
|
|
148
|
64
|
100
|
|
|
|
379
|
$changehsh{$coda} = $posarr[0] if (scalar @posarr); |
149
|
|
|
|
|
|
|
} |
150
|
|
|
|
|
|
|
} |
151
|
|
|
|
|
|
|
|
152
|
1
|
|
|
|
|
4
|
my $offset = 0; |
153
|
1
|
|
|
|
|
2
|
my $newseq = q{}; |
154
|
1
|
|
|
|
|
6
|
while ($offset < length($nucseq)) |
155
|
|
|
|
|
|
|
{ |
156
|
200
|
|
|
|
|
226
|
my $curcod = substr($nucseq, $offset, 3); |
157
|
200
|
|
|
|
|
225
|
$newseq .= $changehsh{$curcod}; |
158
|
200
|
|
|
|
|
281
|
$offset += 3; |
159
|
|
|
|
|
|
|
} |
160
|
1
|
|
|
|
|
7
|
return $newseq; |
161
|
|
|
|
|
|
|
} |
162
|
|
|
|
|
|
|
|
163
|
|
|
|
|
|
|
=head2 _codonJuggle_most_different_sequence |
164
|
|
|
|
|
|
|
|
165
|
|
|
|
|
|
|
=cut |
166
|
|
|
|
|
|
|
|
167
|
|
|
|
|
|
|
sub _codonJuggle_most_different_sequence |
168
|
|
|
|
|
|
|
{ |
169
|
1
|
|
|
1
|
|
3
|
my ($codon_table, $reverse_codon_table, $rscu_table, $nucseq) = @_; |
170
|
1
|
|
|
|
|
2
|
my %changehsh = (); |
171
|
1
|
|
|
|
|
9
|
foreach my $aa (sort keys %$reverse_codon_table) |
172
|
|
|
|
|
|
|
{ |
173
|
21
|
|
|
|
|
28
|
foreach my $coda (@{$reverse_codon_table->{$aa}}) |
|
21
|
|
|
|
|
31
|
|
174
|
|
|
|
|
|
|
{ |
175
|
64
|
|
|
|
|
77
|
my %hsh; |
176
|
64
|
|
|
|
|
81
|
foreach my $codb (@{$reverse_codon_table->{$aa}}) |
|
64
|
|
|
|
|
102
|
|
177
|
|
|
|
|
|
|
{ |
178
|
244
|
|
|
|
|
378
|
$hsh{$codb} = _compare_sequences($coda, $codb); |
179
|
|
|
|
|
|
|
} |
180
|
64
|
|
|
|
|
159
|
my @mdcod = sort {$hsh{$b}->{D} <=> $hsh{$a}->{D} |
181
|
|
|
|
|
|
|
|| $hsh{$b}->{V} <=> $hsh{$a}->{V} |
182
|
|
|
|
|
|
|
|| abs($rscu_table->{$a} - $rscu_table->{$coda}) |
183
|
299
|
50
|
100
|
|
|
749
|
<=> abs($rscu_table->{$b} - $rscu_table->{$coda})} |
184
|
|
|
|
|
|
|
keys %hsh; |
185
|
64
|
100
|
|
|
|
127
|
if (scalar @mdcod > 1) |
186
|
|
|
|
|
|
|
{ |
187
|
62
|
50
|
|
|
|
103
|
shift @mdcod if ($mdcod[0] eq $coda); |
188
|
|
|
|
|
|
|
} |
189
|
64
|
|
|
|
|
250
|
$changehsh{$coda} = $mdcod[0]; |
190
|
|
|
|
|
|
|
} |
191
|
|
|
|
|
|
|
} |
192
|
|
|
|
|
|
|
|
193
|
1
|
|
|
|
|
3
|
my $offset = 0; |
194
|
1
|
|
|
|
|
3
|
my $newseq = q{}; |
195
|
1
|
|
|
|
|
4
|
while ($offset < length($nucseq)) |
196
|
|
|
|
|
|
|
{ |
197
|
200
|
|
|
|
|
223
|
my $curcod = substr($nucseq, $offset, 3); |
198
|
200
|
|
|
|
|
235
|
$newseq .= $changehsh{$curcod}; |
199
|
200
|
|
|
|
|
279
|
$offset += 3; |
200
|
|
|
|
|
|
|
} |
201
|
1
|
|
|
|
|
9
|
return $newseq; |
202
|
|
|
|
|
|
|
} |
203
|
|
|
|
|
|
|
|
204
|
|
|
|
|
|
|
|
205
|
|
|
|
|
|
|
=head2 _codonJuggle_random |
206
|
|
|
|
|
|
|
|
207
|
|
|
|
|
|
|
=cut |
208
|
|
|
|
|
|
|
|
209
|
|
|
|
|
|
|
sub _codonJuggle_random |
210
|
|
|
|
|
|
|
{ |
211
|
10000
|
|
|
10000
|
|
15937
|
my ($codon_table, $reverse_codon_table, $rscu_table, $nucseq) = @_; |
212
|
10000
|
|
|
|
|
12957
|
my $cod_highs = {}; |
213
|
10000
|
|
|
|
|
12495
|
foreach my $aa (keys %{$reverse_codon_table}) |
|
10000
|
|
|
|
|
32830
|
|
214
|
|
|
|
|
|
|
{ |
215
|
210000
|
|
|
|
|
272700
|
$cod_highs->{$aa} = []; |
216
|
210000
|
|
|
|
|
217393
|
foreach my $codon (@{$reverse_codon_table->{$aa}}) |
|
210000
|
|
|
|
|
267727
|
|
217
|
|
|
|
|
|
|
{ |
218
|
640000
|
|
|
|
|
643379
|
push @{$cod_highs->{$aa}}, $codon; |
|
640000
|
|
|
|
|
895194
|
|
219
|
|
|
|
|
|
|
} |
220
|
|
|
|
|
|
|
} |
221
|
10000
|
|
|
|
|
17945
|
my $offset = 0; |
222
|
10000
|
|
|
|
|
11610
|
my $newseq = q{}; |
223
|
10000
|
|
|
|
|
17189
|
while ($offset < length($nucseq)) |
224
|
|
|
|
|
|
|
{ |
225
|
50000
|
|
|
|
|
64897
|
my $curcod = substr($nucseq, $offset, 3); |
226
|
50000
|
|
|
|
|
63978
|
my $aa = $codon_table->{$curcod}; |
227
|
50000
|
|
|
|
|
51077
|
my $index = _random_index(scalar @{$cod_highs->{$aa}}); |
|
50000
|
|
|
|
|
88630
|
|
228
|
50000
|
|
|
|
|
80462
|
$newseq .= $cod_highs->{$aa}->[$index]; |
229
|
50000
|
|
|
|
|
80047
|
$offset += 3; |
230
|
|
|
|
|
|
|
} |
231
|
10000
|
|
|
|
|
58595
|
return $newseq; |
232
|
|
|
|
|
|
|
} |
233
|
|
|
|
|
|
|
|
234
|
|
|
|
|
|
|
1; |
235
|
|
|
|
|
|
|
|
236
|
|
|
|
|
|
|
__END__ |