0
|
1 #!@WHICHPYTHON@
|
|
2
|
|
3 import copy, string, sys
|
|
4
|
|
5 #------------------ Alphabet -------------------
|
|
6
|
|
7 class Alphabet(object):
|
|
8 """Biological alphabet class.
|
|
9 This defines the set of symbols from which various objects can be built, e.g. sequences and motifs.
|
|
10 The symbol set is immutable and accessed as a tuple.
|
|
11 symstr: symbols in alphabet as either a tuple or string
|
|
12 complement: dictionary defining letters and their complement
|
|
13 """
|
|
14 def __init__(self, symstr, complement = None):
|
|
15 """Construct an alphabet from a string or tuple of characters.
|
|
16 Lower case characters will be converted to upper case.
|
|
17 An optional mapping for complements may be provided.
|
|
18 Example:
|
|
19 >>> alpha = sequence.Alphabet('ACGTttga', {'A':'C', 'G':'T'})
|
|
20 >>> alpha.getSymbols()
|
|
21 will construct the DNA alphabet and output:
|
|
22 ('A', 'C', 'G', 'T')
|
|
23 """
|
|
24 symlst = []
|
|
25 for s in [str(sym).upper()[0] for sym in symstr]:
|
|
26 if not s in symlst:
|
|
27 symlst.append(s)
|
|
28 self.symbols = tuple(symlst)
|
|
29 if complement != None:
|
|
30 # expand the mapping and check for contradictions
|
|
31 cmap = {}
|
|
32 for s in self.symbols:
|
|
33 c = complement.get(s, None)
|
|
34 if c != None:
|
|
35 if s in cmap and cmap[s] != c:
|
|
36 raise RuntimeError("Alphabet complement map "
|
|
37 "contains contradictory mapping")
|
|
38 cmap[s] = c
|
|
39 cmap[c] = s
|
|
40 # replace mapping with indicies
|
|
41 cimap = {}
|
|
42 for idx in range (len(self.symbols)):
|
|
43 s = self.symbols[idx]
|
|
44 if s in cmap:
|
|
45 cimap[cmap[s]] = idx
|
|
46 # create tuple
|
|
47 cidxlst = []
|
|
48 for idx in range (len(self.symbols)):
|
|
49 cidxlst.append(cimap.get(self.symbols[idx], None))
|
|
50 self.complements = tuple(cidxlst)
|
|
51 else:
|
|
52 self.complements = None
|
|
53
|
|
54 def getSymbols(self):
|
|
55 """Retrieve a tuple with all symbols, immutable membership and order"""
|
|
56 return self.symbols
|
|
57
|
|
58 def getComplements(self):
|
|
59 """Retrieve a tuple with all complement indicies, immutable"""
|
|
60 return self.complements
|
|
61
|
|
62 def isValidSymbol(self, sym):
|
|
63 """Check if the symbol is a member of alphabet"""
|
|
64 return any([s==sym for s in self.symbols])
|
|
65
|
|
66 def getIndex(self, sym):
|
|
67 """Retrieve the index of the symbol (immutable)"""
|
|
68 for idx in range (len(self.symbols)):
|
|
69 if self.symbols[idx] == sym:
|
|
70 return idx
|
|
71 raise RuntimeError("Symbol " + sym + " does not exist in alphabet")
|
|
72
|
|
73 def isComplementable(self):
|
|
74 return self.complements != None
|
|
75
|
|
76 def getComplement(self, sym):
|
|
77 """Retrieve the complement of the symbol (immutable)"""
|
|
78 return self.symbols[self.complements[self.getIndex(sym)]];
|
|
79
|
|
80 def isValidString(self, symstr):
|
|
81 """Check if the string contains only symbols that belong to the alphabet"""
|
|
82 found = True
|
|
83 for sym in symstr:
|
|
84 if self.isValidSymbol(sym) == False:
|
|
85 return False
|
|
86 return True
|
|
87
|
|
88 def getLen(self):
|
|
89 """Retrieve the number of symbols in (the length of) the alphabet"""
|
|
90 return len(self.symbols)
|
|
91
|
|
92 # pre-defined alphabets that can be specified by their name
|
|
93 predefAlphabets = [
|
|
94 ("DNA" , Alphabet('ACGT', {'A':'T', 'G':'C'})),
|
|
95 ("RNA" , Alphabet('ACGU')),
|
|
96 ("Extended DNA" , Alphabet('ACGTYRN')),
|
|
97 ("Protein" , Alphabet('ACDEFGHIKLMNPQRSTVWY')),
|
|
98 ("Extended Protein" , Alphabet('ACDEFGHIKLMNPQRSTVWYX')),
|
|
99 ("TM Labels" , Alphabet('MIO'))
|
|
100 ]
|
|
101
|
|
102 def getAlphabet(name):
|
|
103 """Retrieve a pre-defined alphabet by name.
|
|
104 Currently, "Protein", "DNA", "RNA", "Extended DNA", "Extended Protein" and "TM Labels" are available.
|
|
105 Example:
|
|
106 >>> alpha = sequence.getAlphabet('Protein')
|
|
107 >>> alpha.getSymbols()
|
|
108 will retrieve the 20 amino acid alphabet and output the tuple:
|
|
109 ('A', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'V', 'W', 'Y')
|
|
110 """
|
|
111 for (xname, xalpha) in predefAlphabets:
|
|
112 if xname == name:
|
|
113 return xalpha
|
|
114 return None
|
|
115
|
|
116 #------------------ Sequence -------------------
|
|
117
|
|
118 class Sequence(object):
|
|
119 """Biological sequence class. Sequence data is immutable.
|
|
120
|
|
121 data: the sequence data as a tuple or string
|
|
122 alpha: the alphabet from which symbols are taken
|
|
123 name: the sequence name, if any
|
|
124 info: can contain additional sequence information apart from the name
|
|
125 """
|
|
126 def __init__(self, sequence, alpha = None, name = "", seqinfo = ""):
|
|
127 """Create a sequence with sequence data.
|
|
128 Specifying the alphabet is optional, so is the name and info.
|
|
129 Example:
|
|
130 >>> myseq = sequence.Sequence('MVSAKKVPAIAMSFGVSF')
|
|
131 will create a sequence with name "", and assign one of the predefined alphabets on basis of what symbols were used.
|
|
132 >>> myseq.getAlphabet().getSymbols()
|
|
133 will most likely output the standard protein alphabet:
|
|
134 ('A', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'V', 'W', 'Y')
|
|
135 """
|
|
136 if type(sequence) is str:
|
|
137 self.data = tuple(sequence.upper())
|
|
138 elif type(sequence) is tuple:
|
|
139 self.data = sequence
|
|
140 elif type(sequence) is list:
|
|
141 self.data = tuple([s.upper() for s in sequence])
|
|
142 else:
|
|
143 raise RuntimeError("Sequence data is not specified correctly: must be string or tuple")
|
|
144 # Resolve choice of alphabet
|
|
145 validAlphabet = False
|
|
146 if alpha == None: # Alphabet is not set, attempt to set it automatically...
|
|
147 for (xname, xalpha) in predefAlphabets: # Iterate through each predefined alphabet, in order
|
|
148 if xalpha.isValidString( self.data ): # This alphabet works, go with it
|
|
149 self.alpha = alpha = xalpha
|
|
150 validAlphabet = True
|
|
151 break
|
|
152 self.name = name
|
|
153 self.info = seqinfo
|
|
154 if validAlphabet == False: # we were either unsuccessful above or the alphabet was specified so test it
|
|
155 if type(alpha) is str: # check if name is a predefined alphabet
|
|
156 for (xname, xalpha) in predefAlphabets: # Iterate through each predefined alphabet, check for name
|
|
157 if (xname == alpha):
|
|
158 alpha = xalpha
|
|
159 break
|
|
160 if type(alpha) is Alphabet: # the alphabet is specified
|
|
161 if alpha.isValidString(self.data) == False:
|
|
162 raise RuntimeError("Invalid alphabet specified: "+"".join(alpha.getSymbols())+" is not compatible with sequence '"+"".join(self.data)+"'")
|
|
163 else:
|
|
164 self.alpha = alpha
|
|
165 else:
|
|
166 raise RuntimeError("Could not identify alphabet from sequence")
|
|
167
|
|
168 #basic getters and setters for the class
|
|
169 def getName(self):
|
|
170 """Get the name of the sequence"""
|
|
171 return self.name
|
|
172 def getInfo(self):
|
|
173 """Get additional info of the sequence (e.g. from the defline in a FASTA file)"""
|
|
174 return self.info
|
|
175 def getAlphabet(self):
|
|
176 """Retrieve the alphabet that is assigned to this sequence"""
|
|
177 return self.alpha
|
|
178 def setName(self, name):
|
|
179 """Change the name of the sequence"""
|
|
180 self.name = name
|
|
181 def setAlphabet(self, alpha):
|
|
182 """Set the alphabet, throws an exception if it is not compatible with the sequence data"""
|
|
183 if type(alpha) is Alphabet:
|
|
184 if alpha.isValid( sequence ) == False:
|
|
185 raise RuntimeError( "Invalid alphabet specified" )
|
|
186 #sequence functions
|
|
187 def getSequence(self):
|
|
188 """Retrieve the sequence data (a tuple of symbols)"""
|
|
189 return self.data
|
|
190 def getString(self):
|
|
191 """Retrieve the sequence data as a string (copy of actual data)"""
|
|
192 return "".join(self.data)
|
|
193 def getLen(self):
|
|
194 """Get the length of the sequence (number of symbols)"""
|
|
195 return len(self.data)
|
|
196 def getSite(self, position, length = 1):
|
|
197 """Retrieve a site in the sequence of desired length.
|
|
198 Note that positions go from 0 to length-1, and that if the requested site
|
|
199 extends beyond those the method throws an exception.
|
|
200 """
|
|
201 if position >= 0 and position <= self.getLen() - length:
|
|
202 if length == 1:
|
|
203 return self.data[position]
|
|
204 else:
|
|
205 return self.data[position:position+length]
|
|
206 else:
|
|
207 raise RuntimeError( "Attempt to access invalid position in sequence "+self.getName() )
|
|
208
|
|
209 def nice(self):
|
|
210 """ A short description of the sequence """
|
|
211 print self.getName(), ":", self.getLen()
|
|
212
|
|
213 def readStrings(filename):
|
|
214 """ Read one or more lines of text from a file--for example an alignment.
|
|
215 Return as a list of strings.
|
|
216 filename: name of file
|
|
217 """
|
|
218 txtlist = []
|
|
219 f = open(filename)
|
|
220 for line in f.readlines():
|
|
221 txtlist.extend(line.split())
|
|
222 return txtlist
|
|
223
|
|
224 def readFASTA(filename, alpha = None):
|
|
225 """ Read one or more sequences from a file in FASTA format.
|
|
226 filename: name of file to load sequences from
|
|
227 alpha: alphabet that is used (if left unspecified, an attempt is made to identify the alphabet for each individual sequence)
|
|
228 """
|
|
229 seqlist = []
|
|
230 seqname = None
|
|
231 seqinfo = None
|
|
232 seqdata = []
|
|
233 fh = open(filename)
|
|
234 thisline = fh.readline()
|
|
235 while (thisline):
|
|
236 if (thisline[0] == '>'): # new sequence
|
|
237 if (seqname): # take care of the data that is already in the buffer before processing the new sequence
|
|
238 try:
|
|
239 seqnew = Sequence(seqdata, alpha, seqname, seqinfo)
|
|
240 seqlist.append(seqnew)
|
|
241 except RuntimeError, e:
|
|
242 print >> sys.stderr, "Warning: "+seqname+" is invalid (ignored): ", e
|
|
243 seqinfo = thisline[1:-1] # everything on the defline is "info"
|
|
244 seqname = seqinfo.split()[0] # up to first space
|
|
245 seqdata = []
|
|
246 else: # pull out the sequence data
|
|
247 cleanline = thisline.split()
|
|
248 for line in cleanline:
|
|
249 seqdata.extend(tuple(line.strip('*'))) # sometimes a line ends with an asterisk in FASTA files
|
|
250 thisline = fh.readline()
|
|
251
|
|
252 if (seqname):
|
|
253 try:
|
|
254 seqnew = Sequence(seqdata, alpha, seqname, seqinfo)
|
|
255 seqlist.append(seqnew)
|
|
256 except RuntimeError, e:
|
|
257 print >> sys.stderr, "Warning: " + seqname + " is invalid (ignored): ", e
|
|
258 else:
|
|
259 raise RuntimeError("No sequences on FASTA format found in this file")
|
|
260 fh.close()
|
|
261 return seqlist
|
|
262
|
|
263 def _writeOneFASTA(sequence, filehandle):
|
|
264 """Write one sequence in FASTA format to an already open file"""
|
|
265 filehandle.write(">" + sequence.getName()+"\n")
|
|
266 data = sequence.getSequence()
|
|
267 lines = ( sequence.getLen() - 1) / 60 + 1
|
|
268 for i in range(lines):
|
|
269 #note: python lets us get the last line (var length) free
|
|
270 #lineofseq = data[i*60 : (i+1)*60] + "\n"
|
|
271 lineofseq = "".join(data[i*60 : (i+1)*60]) + "\n"
|
|
272 filehandle.write(lineofseq)
|
|
273
|
|
274 def writeFASTA(sequence, filename):
|
|
275 """Write a list (or a single) of sequences to a file in the FASTA format"""
|
|
276 fh = open(filename, "w")
|
|
277 if isinstance(sequence, Sequence):
|
|
278 _writeOneFASTA(sequence, fh)
|
|
279 else:
|
|
280 for seq in sequence:
|
|
281 if isinstance(seq, Sequence):
|
|
282 _writeOneFASTA(seq, fh)
|
|
283 else:
|
|
284 print >> sys.stderr, "Warning: could not write " + seq.getName() + " (ignored)."
|
|
285 fh.flush()
|
|
286 fh.close()
|
|
287
|
|
288 #------------------ Distrib -------------------
|
|
289
|
|
290 class Distrib(object):
|
|
291 """Class for storing a multinomial probability distribution over the symbols in an alphabet"""
|
|
292 def __init__(self, alpha, pseudo_count = 0.0):
|
|
293 self.alpha = alpha
|
|
294 self.tot = pseudo_count * self.alpha.getLen()
|
|
295 self.cnt = [pseudo_count for _ in range( self.alpha.getLen() )]
|
|
296
|
|
297 def __deepcopy__(self, memo):
|
|
298 dup = Distrib(self.alpha)
|
|
299 dup.tot = copy.deepcopy(self.tot, memo)
|
|
300 dup.cnt = copy.deepcopy(self.cnt, memo)
|
|
301 return dup
|
|
302
|
|
303 def count(self, syms = None ):
|
|
304 """Count an observation of a symbol"""
|
|
305 if syms == None:
|
|
306 syms = self.alpha.getSymbols()
|
|
307 for sym in syms:
|
|
308 idx = self.alpha.getIndex( sym )
|
|
309 self.cnt[idx] += 1.0
|
|
310 self.tot += 1
|
|
311
|
|
312 def complement(self):
|
|
313 """Complement the counts, throw an error if this is impossible"""
|
|
314 if not self.alpha.isComplementable():
|
|
315 raise RuntimeError("Attempt to complement a Distrib "
|
|
316 "based on a non-complementable alphabet.")
|
|
317 coms = self.alpha.getComplements()
|
|
318 new_count = []
|
|
319 for idx in range(len(coms)):
|
|
320 cidx = coms[idx]
|
|
321 if cidx == None:
|
|
322 cidx = idx
|
|
323 new_count.append(self.cnt[cidx])
|
|
324 self.cnt = new_count
|
|
325 return self
|
|
326
|
|
327 def reset(self):
|
|
328 """Reset the distribution, that is, restart counting."""
|
|
329 self.tot = 0
|
|
330 self.cnt = [0.0 for _ in range( self.alpha.getLen() )]
|
|
331
|
|
332 def getFreq(self, sym = None):
|
|
333 """Determine the probability distribution from the current counts.
|
|
334 The order in which probabilities are given follow the order of the symbols in the alphabet."""
|
|
335 if self.tot > 0:
|
|
336 if sym == None:
|
|
337 freq = tuple([ y / self.tot for y in self.cnt ])
|
|
338 return freq
|
|
339 else:
|
|
340 idx = self.alpha.getIndex( sym )
|
|
341 return self.cnt[idx] / self.tot
|
|
342 return None
|
|
343
|
|
344 def pretty(self):
|
|
345 """Retrieve the probabilites for all symbols and return as a pretty table (a list of text strings)"""
|
|
346 table = ["".join(["%4s " % s for s in self.alpha.getSymbols()])]
|
|
347 table.append("".join(["%3.2f " % y for y in Distrib.getFreq(self)]))
|
|
348 return table
|
|
349
|
|
350 def getSymbols(self):
|
|
351 """Get the symbols in the alphabet in the same order as probabilities are given."""
|
|
352 return self.alpha.getSymbols()
|
|
353
|
|
354 def getAlphabet(self):
|
|
355 """Get the alphabet over which the distribution is defined."""
|
|
356 return self.alpha
|
|
357
|
|
358 #------------------ Motif (and subclasses) -------------------
|
|
359
|
|
360 class Motif(object):
|
|
361 """ Sequence motif class--defining a pattern that can be searched in sequences.
|
|
362 This class is not intended for direct use. Instead use and develop sub-classes (see below).
|
|
363 """
|
|
364 def __init__(self, alpha):
|
|
365 self.len = 0
|
|
366 self.alpha = alpha
|
|
367
|
|
368 def getLen(self):
|
|
369 """Get the length of the motif"""
|
|
370 return self.len
|
|
371
|
|
372 def getAlphabet(self):
|
|
373 """Get the alphabet that is used in the motif"""
|
|
374 return self.alpha
|
|
375
|
|
376 def isAlphabet(self, seqstr):
|
|
377 """Check if the sequence can be processed by this motif"""
|
|
378 mystr = seqstr
|
|
379 if type(seqstr) is Sequence:
|
|
380 mystr = seqstr.getString()
|
|
381 return self.getAlphabet().isValidString(mystr)
|
|
382
|
|
383 import re
|
|
384
|
|
385 class RegExp(Motif):
|
|
386 """A motif class that defines the pattern in terms of a regular expression"""
|
|
387 def __init__(self, alpha, re_string):
|
|
388 Motif.__init__(self, alpha)
|
|
389 self.pattern = re.compile(re_string)
|
|
390
|
|
391 def match(self, seq):
|
|
392 """Find matches to the motif in a specified sequence.
|
|
393 The method is a generator, hence subsequent hits can be retrieved using next().
|
|
394 The returned result is a tuple (position, match-sequence, score), where score is
|
|
395 always 1.0 since a regular expression is either true or false (not returned).
|
|
396 """
|
|
397 myseq = seq
|
|
398 if not type(seq) is Sequence:
|
|
399 myseq = Sequence(seq, self.alpha)
|
|
400 mystr = myseq.getString()
|
|
401 if not Motif.isAlphabet(self, mystr):
|
|
402 raise RuntimeError("Motif alphabet is not valid for sequence " + myseq.getName())
|
|
403 for m in re.finditer(self.pattern, mystr):
|
|
404 yield (m.start(), m.group(), 1.0)
|
|
405
|
|
406 import math, time
|
|
407
|
|
408 # Variables used by the PWM for creating an EPS file
|
|
409 _colour_def = (
|
|
410 "/black [0 0 0] def\n"
|
|
411 "/red [0.8 0 0] def\n"
|
|
412 "/green [0 0.5 0] def\n"
|
|
413 "/blue [0 0 0.8] def\n"
|
|
414 "/yellow [1 1 0] def\n"
|
|
415 "/purple [0.8 0 0.8] def\n"
|
|
416 "/magenta [1.0 0 1.0] def\n"
|
|
417 "/cyan [0 1.0 1.0] def\n"
|
|
418 "/pink [1.0 0.8 0.8] def\n"
|
|
419 "/turquoise [0.2 0.9 0.8] def\n"
|
|
420 "/orange [1 0.7 0] def\n"
|
|
421 "/lightred [0.8 0.56 0.56] def\n"
|
|
422 "/lightgreen [0.35 0.5 0.35] def\n"
|
|
423 "/lightblue [0.56 0.56 0.8] def\n"
|
|
424 "/lightyellow [1 1 0.71] def\n"
|
|
425 "/lightpurple [0.8 0.56 0.8] def\n"
|
|
426 "/lightmagenta [1.0 0.7 1.0] def\n"
|
|
427 "/lightcyan [0.7 1.0 1.0] def\n"
|
|
428 "/lightpink [1.0 0.9 0.9] def\n"
|
|
429 "/lightturquoise [0.81 0.9 0.89] def\n"
|
|
430 "/lightorange [1 0.91 0.7] def\n")
|
|
431 _colour_dict = (
|
|
432 "/fullColourDict <<\n"
|
|
433 " (G) orange\n"
|
|
434 " (T) green\n"
|
|
435 " (C) blue\n"
|
|
436 " (A) red\n"
|
|
437 " (U) green\n"
|
|
438 ">> def\n"
|
|
439 "/mutedColourDict <<\n"
|
|
440 " (G) lightorange\n"
|
|
441 " (T) lightgreen\n"
|
|
442 " (C) lightblue\n"
|
|
443 " (A) lightred\n"
|
|
444 " (U) lightgreen\n"
|
|
445 ">> def\n"
|
|
446 "/colorDict fullColourDict def\n")
|
|
447
|
|
448 _eps_defaults = {
|
|
449 'LOGOTYPE': 'NA',
|
|
450 'FONTSIZE': '12',
|
|
451 'TITLEFONTSIZE': '12',
|
|
452 'SMALLFONTSIZE': '6',
|
|
453 'TOPMARGIN': '0.9',
|
|
454 'BOTTOMMARGIN': '0.9',
|
|
455 'YAXIS': 'true',
|
|
456 'YAXISLABEL': 'bits',
|
|
457 'XAXISLABEL': '',
|
|
458 'TITLE': '',
|
|
459 'ERRORBARFRACTION': '1.0',
|
|
460 'SHOWINGBOX': 'false',
|
|
461 'BARBITS': '2.0',
|
|
462 'TICBITS': '1',
|
|
463 'COLORDEF': _colour_def,
|
|
464 'COLORDICT': _colour_dict,
|
|
465 'SHOWENDS': 'false',
|
|
466 'NUMBERING': 'true',
|
|
467 'OUTLINE': 'false',
|
|
468 }
|
|
469 class PWM(Motif):
|
|
470 """This motif subclass defines a pattern in terms of a position weight matrix.
|
|
471 An alphabet must be provided. A pseudo-count to be added to each count is
|
|
472 optional. A uniform background distribution is used by default.
|
|
473 """
|
|
474 def __init__(self, alpha):
|
|
475 Motif.__init__(self, alpha) # set alphabet of this multinomial distribution
|
|
476 self.background = Distrib(alpha) # the default background ...
|
|
477 self.background.count(alpha.getSymbols()) # ... is uniform
|
|
478 self.nsites = 0
|
|
479
|
|
480 def setFromAlignment(self, aligned, pseudo_count = 0.0):
|
|
481 """Set the probabilities in the PWM from an alignment.
|
|
482 The alignment is a list of equal-length strings (see readStrings), OR
|
|
483 a list of Sequence.
|
|
484 """
|
|
485 self.cols = -1
|
|
486 self.nsites = len(aligned)
|
|
487 seqs = []
|
|
488 # Below we create a list of Sequence from the alignment,
|
|
489 # while doing some error checking, and figure out the number of columns
|
|
490 for s in aligned:
|
|
491 # probably a text string, so we make a nameless sequence from it
|
|
492 if not type(s) is Sequence:
|
|
493 s=Sequence(s, Motif.getAlphabet(self))
|
|
494 else:
|
|
495 # it was a sequence, so we check that the alphabet in
|
|
496 # this motif will be able to process it
|
|
497 if not Motif.isAlphabet(self, s):
|
|
498 raise RuntimeError("Motif alphabet is not valid for sequence " + s.getName())
|
|
499 if self.cols == -1:
|
|
500 self.cols = s.getLen()
|
|
501 elif self.cols != s.getLen():
|
|
502 raise RuntimeError("Sequences in alignment are not of equal length")
|
|
503 seqs.append(s)
|
|
504 # The line below initializes the list of Distrib (one for each column of the alignment)
|
|
505 self.counts = [Distrib(Motif.getAlphabet(self), pseudo_count) for _ in range(self.cols)]
|
|
506 # Next, we do the counting, column by column
|
|
507 for c in range( self.cols ): # iterate through columns
|
|
508 for s in seqs: # iterate through rows
|
|
509 # determine the index of the symbol we find at this position (row, column c)
|
|
510 self.counts[c].count(s.getSite(c))
|
|
511 # Update the length
|
|
512 self.len = self.cols
|
|
513
|
|
514 def reverseComplement(self):
|
|
515 """Reverse complement the PWM"""
|
|
516 i = 0
|
|
517 j = len(self.counts)-1
|
|
518 while (i < j):
|
|
519 temp = self.counts[i];
|
|
520 self.counts[i] = self.counts[j]
|
|
521 self.counts[j] = temp
|
|
522 self.counts[i].complement()
|
|
523 self.counts[j].complement()
|
|
524 i += 1;
|
|
525 j -= 1;
|
|
526 if i == j:
|
|
527 self.counts[i].complement()
|
|
528 return self
|
|
529
|
|
530 def getNSites(self):
|
|
531 """Get the number of sites that made the PWM"""
|
|
532 return self.nsites
|
|
533
|
|
534 def setBackground(self, distrib):
|
|
535 """Set the background distribution"""
|
|
536 if not distrib.getAlphabet() == Motif.getAlphabet(self):
|
|
537 raise RuntimeError("Incompatible alphabets")
|
|
538 self.background = distrib
|
|
539
|
|
540 def getFreq(self, col = None, sym = None):
|
|
541 """Get the probabilities for all positions in the PWM (a list of Distribs)"""
|
|
542 if (col == None):
|
|
543 return [y.getFreq() for y in self.counts]
|
|
544 else:
|
|
545 return self.counts[col].getFreq(sym)
|
|
546
|
|
547 def pretty(self):
|
|
548 """Retrieve the probabilites for all positions in the PWM as a pretty table (a list of text strings)"""
|
|
549 #table = ["".join(["%8s " % s for s in self.alpha.getSymbols()])]
|
|
550 table = []
|
|
551 for row in PWM.getFreq(self):
|
|
552 table.append("".join(["%8.6f " % y for y in row]))
|
|
553 return table
|
|
554
|
|
555 def logoddsPretty(self, bkg):
|
|
556 """Retrieve the (base-2) log-odds for all positions in the PWM as a pretty table (a list of text strings)"""
|
|
557 table = []
|
|
558 for row in PWM.getFreq(self):
|
|
559 #table.append("".join(["%8.6f " % (math.log((row[i]+1e-6)/bkg[i])/math.log(2)) for i in range(len(row))]))
|
|
560 table.append("".join(["%8.6f " % (math.log((row[i])/bkg[i])/math.log(2)) for i in range(len(row))]))
|
|
561 #table.append("".join(["%8.6f " % row[i] for i in range(len(row))]))
|
|
562 return table
|
|
563
|
|
564
|
|
565 def consensus_sequence(self):
|
|
566 """
|
|
567 Get the consensus sequence corresponding to a PWM.
|
|
568 Consensus sequence is the letter in each column
|
|
569 with the highest probability.
|
|
570 """
|
|
571 consensus = ""
|
|
572 alphabet = Motif.getAlphabet(self).getSymbols()
|
|
573 for pos in range(self.cols):
|
|
574 best_letter = alphabet[0]
|
|
575 best_p = self.counts[pos].getFreq(best_letter)
|
|
576 for letter in alphabet[1:]:
|
|
577 p = self.counts[pos].getFreq(letter)
|
|
578 if p > best_p:
|
|
579 best_p = p
|
|
580 best_letter = letter
|
|
581 consensus += best_letter
|
|
582 return consensus
|
|
583
|
|
584
|
|
585 def consensus(self):
|
|
586 """
|
|
587 Get the consensus corresponding to a PWM.
|
|
588 Consensus at each column of motif is a list of
|
|
589 characters with non-zero probabilities.
|
|
590 """
|
|
591 consensus = []
|
|
592 for pos in range(self.cols):
|
|
593 matches = []
|
|
594 for letter in Motif.getAlphabet(self).getSymbols():
|
|
595 p = self.counts[pos].getFreq(letter)
|
|
596 if p > 0:
|
|
597 matches += letter
|
|
598 consensus.append(matches)
|
|
599 return consensus
|
|
600
|
|
601
|
|
602 def getScore(self, seq, start):
|
|
603 """Score this particular list of symbols using the PFM (background needs to be set separately)"""
|
|
604 sum = 0.0
|
|
605 seqdata = seq.getSequence()[start : start+self.cols]
|
|
606 for pos in range(len(seqdata)):
|
|
607 q = self.counts[pos].getFreq(seqdata[pos])
|
|
608 if q == 0:
|
|
609 q = 0.0001 # to avoid log(0) == -Infinity
|
|
610 logodds = math.log(q / self.background.getFreq(seqdata[pos]))
|
|
611 sum += logodds
|
|
612 return sum
|
|
613
|
|
614 def match(self, seq, _LOG0 = -10):
|
|
615 """Find matches to the motif in a specified sequence.
|
|
616 The method is a generator, hence subsequent hits can be retrieved using next().
|
|
617 The returned result is a tuple (position, match-sequence, score).
|
|
618 The optional parameter _LOG0 specifies a lower bound on reported logodds scores.
|
|
619 """
|
|
620 myseq = seq
|
|
621 if not type(seq) is Sequence:
|
|
622 myseq = Sequence(seq, self.alpha)
|
|
623 if not Motif.isAlphabet(self, myseq):
|
|
624 raise RuntimeError("Motif alphabet is not valid for sequence " + myseq.getName())
|
|
625 for pos in range(myseq.getLen() - self.cols):
|
|
626 score = PWM.getScore(self, myseq, pos)
|
|
627 if score > _LOG0:
|
|
628 yield (pos, "".join(myseq.getSite(pos, self.cols)), score)
|
|
629
|
|
630 def writeEPS(self, program, template_file, eps_fh,
|
|
631 timestamp = time.localtime()):
|
|
632 """Write out a DNA motif to EPS format."""
|
|
633 small_dfmt = "%d.%m.%Y %H:%M"
|
|
634 full_dfmt = "%d.%m.%Y %H:%M:%S %Z"
|
|
635 small_date = time.strftime(small_dfmt, timestamp)
|
|
636 full_date = time.strftime(full_dfmt, timestamp)
|
|
637 points_per_cm = 72.0 / 2.54
|
|
638 height = 4.5
|
|
639 width = self.getLen() * 0.8 + 2
|
|
640 width = min(30, width)
|
|
641 points_height = int(height * points_per_cm)
|
|
642 points_width = int(width * points_per_cm)
|
|
643 defaults = _eps_defaults.copy()
|
|
644 defaults['CREATOR'] = program
|
|
645 defaults['CREATIONDATE'] = full_date
|
|
646 defaults['LOGOHEIGHT'] = str(height)
|
|
647 defaults['LOGOWIDTH'] = str(width)
|
|
648 defaults['FINEPRINT'] = program + ' ' + small_date
|
|
649 defaults['CHARSPERLINE'] = str(self.getLen())
|
|
650 defaults['BOUNDINGHEIGHT'] = str(points_height)
|
|
651 defaults['BOUNDINGWIDTH'] = str(points_width)
|
|
652 defaults['LOGOLINEHEIGHT'] = str(height)
|
|
653 with open(template_file, 'r') as template_fh:
|
|
654 m_var = re.compile("\{\$([A-Z]+)\}")
|
|
655 for line in template_fh:
|
|
656 last = 0
|
|
657 match = m_var.search(line)
|
|
658 while (match):
|
|
659 if (last < match.start()):
|
|
660 prev = line[last:match.start()]
|
|
661 eps_fh.write(prev)
|
|
662 key = match.group(1)
|
|
663 if (key == "DATA"):
|
|
664 eps_fh.write("\nStartLine\n")
|
|
665 for pos in range(self.getLen()):
|
|
666 eps_fh.write("({0:d}) startstack\n".format(pos+1))
|
|
667 stack = []
|
|
668 # calculate the stack information content
|
|
669 alpha_ic = 2
|
|
670 h = 0
|
|
671 for sym in self.getAlphabet().getSymbols():
|
|
672 freq = self.getFreq(pos, sym)
|
|
673 if (freq == 0):
|
|
674 continue
|
|
675 h -= (freq * math.log(freq, 2))
|
|
676 stack_ic = alpha_ic - h
|
|
677 # calculate the heights of each symbol
|
|
678 for sym in self.getAlphabet().getSymbols():
|
|
679 freq = self.getFreq(pos, sym)
|
|
680 if (freq == 0):
|
|
681 continue
|
|
682 stack.append((freq * stack_ic, sym))
|
|
683 stack.sort();
|
|
684 # output the symbols
|
|
685 for symh, sym in stack:
|
|
686 eps_fh.write(" {0:f} ({1:s}) numchar\n".format(
|
|
687 symh, sym))
|
|
688 eps_fh.write("endstack\n\n")
|
|
689 eps_fh.write("EndLine\n")
|
|
690 elif (key in defaults):
|
|
691 eps_fh.write(defaults[key])
|
|
692 else:
|
|
693 raise RuntimeError('Unknown variable "' + key +
|
|
694 '" in EPS template')
|
|
695 last = match.end();
|
|
696 match = m_var.search(line, last)
|
|
697 if (last < len(line)):
|
|
698 eps_fh.write(line[last:])
|
|
699
|
|
700
|
|
701 #------------------ Main method -------------------
|
|
702 # Executed if you run this file from the operating system prompt, e.g.
|
|
703 # > python sequence.py
|
|
704
|
|
705 if __name__=='__main__':
|
|
706 alpha = getAlphabet('Extended DNA')
|
|
707 #seqs = readFASTA('pos.fasta')
|
|
708 seqs = []
|
|
709 aln = readStrings('tmp0')
|
|
710 #regexp = RegExp(alpha, '[AG]G.[DE]TT[AS].')
|
|
711 pwm = PWM(alpha)
|
|
712 pwm.setFromAlignment(aln)
|
|
713 for row in pwm.pretty():
|
|
714 print row
|
|
715 for s in seqs:
|
|
716 print s.getName(), s.getLen(), s.getAlphabet().getSymbols()
|
|
717 for m in regexp.match( s ):
|
|
718 print "pos: %d pat: %s %4.2f" % (m[0], m[1], m[2])
|
|
719 for m in pwm.match( s ):
|
|
720 print "pos: %d pat: %s %4.2f" % (m[0], m[1], m[2])
|