Delete unneeded directory
[jabaws.git] / website / archive / binaries / mac / src / globplot / biopython-1.50 / Bio / utils.py
diff --git a/website/archive/binaries/mac/src/globplot/biopython-1.50/Bio/utils.py b/website/archive/binaries/mac/src/globplot/biopython-1.50/Bio/utils.py
deleted file mode 100644 (file)
index bf0d317..0000000
+++ /dev/null
@@ -1,163 +0,0 @@
-# Copyright 2000 by Andrew Dalke.
-# All rights reserved.
-# This code is part of the Biopython distribution and governed by its
-# license.  Please see the LICENSE file that should have been included
-# as part of this package.
-
-"""Miscellaneous functions for dealing with sequences (obsolete?)."""
-
-import Seq
-import Alphabet
-
-from PropertyManager import default_manager
-
-def translate(seq, id = None):
-    """Translate a sequence (DEPRECATED)."""
-    import warnings
-    warnings.warn("Bio.utils.translate() has been deprecated, and we" \
-                  +" intend to remove it in a future release of Biopython."\
-                  +" Please use the translate method or function in Bio.Seq"\
-                  +" instead, as described in the Tutorial.",
-                  DeprecationWarning)
-    if id is None:
-        s = "translator"
-    else:
-        s = "translator.id.%d" % id
-    translator = default_manager.resolve(seq.alphabet, s)
-    return translator.translate(seq)
-
-def translate_to_stop(seq, id = None):
-    """Translate a sequence up to the first in frame stop codon (DEPRECATED)."""
-    import warnings
-    warnings.warn("Bio.utils.translate_to_stop() has been deprecated, and we" \
-                  +" intend to remove it in a future release of Biopython."\
-                  +" Please use the translate method or function in Bio.Seq"\
-                  +" instead, as described in the Tutorial.",
-                  DeprecationWarning)
-    if id is None:
-        s = "translator"
-    else:
-        s = "translator.id.%d" % id
-    translator = default_manager.resolve(seq.alphabet, s)
-    return translator.translate_to_stop(seq)
-
-def back_translate(seq, id = None):
-    """Back-translate a sequence (DEPRECATED)."""
-    import warnings
-    warnings.warn("Bio.utils.back_translate() has been deprecated, and we" \
-                  +" intend to remove it in a future release of Biopython."\
-                  +" If you use it, please tell us on the mailing list.",
-                  DeprecationWarning)
-    if id is None:
-        s = "translator"
-    else:
-        s = "translator.id.%d" % id
-    translator = default_manager.resolve(seq.alphabet, s)
-    return translator.back_translate(seq)
-
-
-def transcribe(seq):
-    """Transcribe a sequence (DEPRECATED)."""
-    import warnings
-    warnings.warn("Bio.utils.transcribe() has been deprecated, and we" \
-                  +" intend to remove it in a future release of Biopython."\
-                  +" Please use the transcribe method or function in"\
-                  +" Bio.Seq instead, as described in the Tutorial.",
-                  DeprecationWarning)
-    transcriber = default_manager.resolve(seq.alphabet, "transcriber")
-    return transcriber.transcribe(seq)
-
-def back_transcribe(seq):
-    """Back-transcribe a sequence (DEPRECATED)."""
-    import warnings
-    warnings.warn("Bio.utils.back_transcribe() has been deprecated, and we" \
-                  +" intend to remove it in a future release of Biopython."\
-                  +" Please use the back_transcribe method or function in"\
-                  +" Bio.Seq instead, as described in the Tutorial.",
-                  DeprecationWarning)
-    transcriber = default_manager.resolve(seq.alphabet, "transcriber")
-    return transcriber.back_transcribe(seq)
-
-def ungap(seq):
-    """given a sequence with gap encoding, return the ungapped sequence"""
-    #TODO - Fix this?  It currently assumes the outmost AlphabetEncoder
-    #is for the gap.  Consider HasStopCodon(Gapped(Protein())) as a test case.
-    gap = seq.gap_char
-    letters = []
-    for c in seq.data:
-        if c != gap:
-            letters.append(c)
-    return Seq.Seq("".join(letters), seq.alphabet.alphabet)
-
-def verify_alphabet(seq):
-    letters = {}
-    for c in seq.alphabet.letters:
-        letters[c] = 1
-    try:
-        for c in seq.data:
-            letters[c]
-    except KeyError:
-        return 0
-    return 1
-
-def count_monomers(seq):
-    dict = {}
-#    bugfix: string.count(s,c) raises an AttributeError. Iddo Friedberg 16 Mar. 04
-#    s = buffer(seq.data)  # works for strings and array.arrays
-    for c in seq.alphabet.letters:
-        dict[c] = seq.data.count(c)
-    return dict
-
-def percent_monomers(seq):
-    dict2 = {}
-    seq_len = len(seq)
-    dict = count_monomers(seq)
-    for m in dict:
-        dict2[m] = dict[m] * 100. / seq_len
-    return dict2
-
-def sum(seq, table, zero = 0.0):
-    total = zero
-    for c in getattr(seq, "data", seq):
-        total = total + table[c]
-    return total
-
-# For ranged addition
-def sum_2ple(seq, table, zero = (0.0, 0.0)):
-    x, y = zero
-    data = getattr(seq, "data", seq)
-    for c in data:
-        x2, y2 = table[c]
-        x = x + x2
-        y = y + y2
-    return (x, y)
-
-def total_weight(seq, weight_table = None):
-    if weight_table is None:
-        weight_table = default_manager.resolve(seq.alphabet, "weight_table")
-    return sum(seq, weight_table)
-
-def total_weight_range(seq, weight_table = None):
-    if weight_table is None:
-        weight_table = default_manager.resolve(seq.alphabet, "weight_range_table")
-    return sum_2ple(seq, weight_table)
-
-def reduce_sequence(seq, reduction_table,new_alphabet=None):
-   """ given an amino-acid sequence, return it in reduced alphabet form based
-       on the letter-translation table passed. Some "standard" tables are in
-       Alphabet.Reduced.
-       seq: a Seq.Seq type sequence
-       reduction_table: a dictionary whose keys are the "from" alphabet, and values
-       are the "to" alphabet"""
-   if new_alphabet is None:
-      new_alphabet = Alphabet.single_letter_alphabet
-      new_alphabet.letters = ''
-      for letter in reduction_table:
-         new_alphabet.letters += letter
-      new_alphabet.size = len(new_alphabet.letters)
-   new_seq = Seq.Seq('',new_alphabet)
-   for letter in seq:
-      new_seq += reduction_table[letter]
-   return new_seq
-
-