1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
|
from math import sqrt
import math
import scipy.stats as s
import array as a
from scipy.optimize import fminbound
from scipy.special import gammaln as gamln
from numpy import log, pi, log10, e, log1p, exp
import numpy as np
from .MultiSiteCommon import MultiSiteCommon, canonicalBaseMap, modNames, ModificationPeakMask, FRAC, FRAClow, FRACup, log10e
from .MixtureEstimationMethods import MixtureEstimationMethods
class ModificationDecode(MultiSiteCommon):
def __init__(self, gbmModel, sequence, rawKinetics, callBounds, methylMinCov, modsToCall=[
'H', 'J', 'K'], methylFractionFlag=False, useLDAFlag=False):
MultiSiteCommon.__init__(self, gbmModel, sequence, rawKinetics)
# Extents that we will attemp to call a modification
self.callStart = callBounds[0]
self.callEnd = callBounds[1]
self.callRange = range(self.callStart, self.callEnd)
self.methylMinCov = methylMinCov
self.modsToCall = modsToCall
self.methylFractionFlag = methylFractionFlag
self.useLDA = useLDAFlag
def decode(self):
"""Use this method to do the full modification finding protocol"""
# Find potential modification sites
self.findAlternates()
# Compute all the required mean ipds under all possible composite
# hypotheses
self.computeContextMeans()
# Fill out the forward matrix
self.fwdRecursion()
# Trace back the fwd matrix and return modification calls
modCalls = self.traceback()
# Compute a confidence for each mod and return results
return self.scoreMods(modCalls)
def findAlternates(self):
""" Use rules about where IPD peaks appear to generate list
the set of possible modified bases that we will test during decoding."""
scoreThresholdLow = 16
scoreThresholdHigh = 19
seq = self.sequence
for (pos, peak) in self.rawKinetics.items():
score = peak['score']
# if self.useLDA:
# Try using LDA model to identify putative Ca5C, regardless of scores
# if peak.has_key('Ca5C'):
# if peak['Ca5C'] < 0:
# self.alternateBases[pos].add('K')
# Exclude points with low score
if score < scoreThresholdLow:
continue
# note -- don't use the tpl in the actual array - use the dict key
# we have reversed the indexing to deal with the reverse strand
if self.callStart <= pos < self.callEnd:
c = seq[pos]
# On-target A peak
if 'H' in self.modsToCall and c == 'A' and score > scoreThresholdHigh:
self.alternateBases[pos].add('H')
# On-target C peak
if 'J' in self.modsToCall and c == 'C' and score > scoreThresholdHigh:
self.alternateBases[pos].add('J')
if 'K' in self.modsToCall:
if c == 'C':
self.alternateBases[pos].add('K')
# peak at -1 or -2 or -6 of a C -- 5caC
if seq[pos - 2] == 'C' and pos - 2 >= self.callStart:
self.alternateBases[pos - 2].add('K')
if seq[pos + 1] == 'C' and pos + 1 < self.callEnd:
self.alternateBases[pos + 1].add('K')
if seq[pos + 2] == 'C' and pos + 2 < self.callEnd:
self.alternateBases[pos + 2].add('K')
if seq[pos + 5] == 'C' and pos + 5 < self.callEnd:
self.alternateBases[pos + 5].add('K')
if seq[pos + 6] == 'C' and pos + 6 < self.callEnd:
self.alternateBases[pos + 6].add('K')
def fwdRecursion(self):
start = self.lStart
end = self.lEnd
# Likelihood of each configuration at each position
scores = dict()
# fwd score matrix and fwd lookback matrix
fwdScore = dict()
fwdPrevState = dict()
# Fill out first column of score & fwd matrix
scores[start] = dict((cfg, self.scorePosition(start, cfg))
for cfg in self.getConfigs(start))
# First column of fwd matrix is same a score matrix, with 'None' in the
# index matrix
fwdScore[start] = scores[start]
fwdPrevState[start] = dict((x, None) for x in scores[start].keys())
for centerPos in range(start + 1, end):
# Score and fwd column for current position
scoreCol = dict()
fwdScoreCol = dict()
fwdPrevStateCol = dict()
# Loop over current state options
for cfg in self.getConfigs(centerPos):
score = self.scorePosition(centerPos, cfg)
scoreCol[cfg] = score
bestPrevState = None
bestScore = -1e20
# Loop over previous state options
for (prevCfg, prevScore) in fwdScore[centerPos - 1].items():
if self.compareStates(
cfg, prevCfg) and prevScore + score > bestScore:
bestScore = prevScore + score
bestPrevState = prevCfg
fwdScoreCol[cfg] = bestScore
fwdPrevStateCol[cfg] = bestPrevState
scores[centerPos] = scoreCol
fwdScore[centerPos] = fwdScoreCol
fwdPrevState[centerPos] = fwdPrevStateCol
self.fwdScore = fwdScore
self.fwdPrevState = fwdPrevState
self.scores = scores
def traceback(self):
"""
Traceback the fwd matrix to get the bset scoring configuration of modifications
"""
start = self.lStart
end = self.lEnd
modCalls = dict()
def cogBase(cfg):
return cfg[self.pre]
pos = end - 1
currentCol = self.fwdScore[end - 1]
bestConfig = max(currentCol, key=lambda x: currentCol[x])
while True:
if cogBase(bestConfig) != self.sequence[pos]:
# Found a modification - save it!
modCalls[pos] = cogBase(bestConfig)
bestConfig = self.fwdPrevState[pos][bestConfig]
pos -= 1
if bestConfig is None:
break
# if self.useLDA:
# allow LDA-predicted sites through to GFF file
# for pos in range(start, end):
# if self.rawKinetics.has_key(pos):
# if self.rawKinetics[pos].has_key('Ca5C'):
# if 'K' in self.modsToCall:
# cutoff = min( 0, self.rawKinetics[pos]['coverage']/20.0 - 3.0 )
# cutoff = 0
# echoSites = [modCalls[pos + i] for i in [-6, -5, -2, -1, 2] if modCalls.has_key(pos + i)]
# else:
# cutoff = -2.25
# echoSites = [modCalls[pos + i] for i in range(-10,11) if modCalls.has_key(pos + i)]
# if self.rawKinetics[pos]['Ca5C'] < cutoff:
# so long as those sites are not in the vicinity of a m6A/m4C call
# if 'H' not in echoSites and 'J' not in echoSites:
# modCalls[pos] = 'K'
# else:
# remove any non-LDA-predicted sites from modCalls dictionary?
# if modCalls.has_key(pos):
# del modCalls[pos]
# correct adjacent calls:
# if self.useLDA:
# for pos in range(start + 2, end - 2, 2 ):
# x = [pos + i for i in range(-2,3) if modCalls.has_key( pos + i) and self.rawKinetics.has_key( pos + i) ]
# y = [modCalls[j] for j in x]
# if y.count('K') > 1:
# tmp = [self.rawKinetics[j]['Ca5C'] for j in x if self.rawKinetics[j].has_key('Ca5C') ]
# if len(tmp) > 0:
# lowest = min(tmp)
# for j in x:
# if self.rawKinetics[j].has_key('Ca5C'):
# if self.rawKinetics[j]['Ca5C'] > lowest:
# del modCalls[j]
#
# if adjacent m5C calls are made by the LDA, select the one that has the lower LDA score (Ca5C)
# for pos in range(start, end):
# if modCalls.has_key(pos) and self.rawKinetics.has_key(pos) and modCalls.has_key(pos+1) and self.rawKinetics.has_key(pos+1):
# if self.rawKinetics[pos].has_key('Ca5C') and self.rawKinetics[pos+1].has_key('Ca5C'):
# if self.rawKinetics[pos]['Ca5C'] < self.rawKinetics[pos+1]['Ca5C']:
# del modCalls[pos+1]
# else:
# del modCalls[pos]
return modCalls
def scoreMods(self, modCalls):
"""
For each modification in the best scoring configuration, score a config excluding the current mod against the winning config
use this value as the Qmod for the deleted modification
"""
qvModCalls = dict()
modSeq = a.array('b')
modSeq.frombytes(bytes(self.sequence, "ascii"))
# Apply the found modifications to the raw sequence
for (pos, call) in modCalls.items():
modSeq[pos] = ord(call)
for (pos, call) in modCalls.items():
# Score the modified template at all positions affected by this mod
modScore = self.scoreRegion(
pos - self.post, pos + self.pre, modSeq)
modScores = self.getRegionScores(
pos - self.post, pos + self.pre, modSeq)
if self.methylFractionFlag and pos in self.rawKinetics:
if self.rawKinetics[pos]["coverage"] > self.methylMinCov:
modifiedMeanVectors = self.getContextMeans(
pos - self.post, pos + self.pre, modSeq)
# Switch back to the unmodified base and re-score
modSeq[pos] = ord(canonicalBaseMap[call])
noModScore = self.scoreRegion(
pos - self.post, pos + self.pre, modSeq)
noModScores = self.getRegionScores(
pos - self.post, pos + self.pre, modSeq)
if self.methylFractionFlag and pos in self.rawKinetics:
if self.rawKinetics[pos]["coverage"] > self.methylMinCov:
unModifiedMeanVectors = self.getContextMeans(
pos - self.post, pos + self.pre, modSeq)
# Put back the modified base
modSeq[pos] = ord(call)
# Compute score difference
llr = modScore - noModScore
# Convert from LLR to phred-scaled probability of modification
qModScore = 10 * llr * log10e + 10 * log1p(exp(-llr)) * log10e
# Figure out which secondary peaks were likely generated by this modification
# What is the posterior that the peak was generated by this mod?
maskPos = self.findMaskPositions(pos, modScores, noModScores)
# FIXME: Without this, currently, the identificationQv score is too low for many Ca5C sites
# if self.useLDA:
# if self.rawKinetics.has_key(pos):
# if self.rawKinetics[pos].has_key('Ca5C'):
# llr = -self.rawKinetics[pos]['Ca5C']
# qModScore = 100 * llr * log10e + 100*log1p(exp(-llr))*log10e
if self.methylFractionFlag and pos in self.rawKinetics:
if self.rawKinetics[pos]["coverage"] > self.methylMinCov:
# Instantiate mixture estimation methods:
mixture = MixtureEstimationMethods(
self.gbmModel.post, self.gbmModel.pre, self.rawKinetics, self.methylMinCov)
# Use modifiedMeanVectors and unmodifiedMeanVectors to
# calculate mixing proportion, and 95% CI limits.
methylFracEst, methylFracLow, methylFracUpp = mixture.estimateMethylatedFractions(
pos, unModifiedMeanVectors, modifiedMeanVectors, ModificationPeakMask[modNames[call]])
qvModCalls[pos] = {'modification': modNames[call], 'QMod': qModScore, 'LLR': llr, 'Mask': maskPos,
FRAC: methylFracEst, FRAClow: methylFracLow, FRACup: methylFracUpp}
else:
qvModCalls[pos] = {'modification': modNames[call],
'QMod': qModScore, 'LLR': llr, 'Mask': maskPos}
else:
# Store the full results
qvModCalls[pos] = {'modification': modNames[call],
'QMod': qModScore, 'LLR': llr, 'Mask': maskPos}
return qvModCalls
def scoreRegion(self, start, end, sequence):
sc = 0.0
for pos in range(start, end + 1):
ctx = sequence[(pos - self.pre):(pos + self.post + 1)
].tobytes().decode("ascii")
if pos in self.scores:
sc += self.scores[pos][ctx]
return sc
def getRegionScores(self, start, end, sequence):
scores = np.zeros(end - start + 1)
for pos in range(start, end + 1):
ctx = sequence[(pos - self.pre):(pos + self.post + 1)
].tobytes().decode("ascii")
if pos in self.scores:
scores[pos - start] = self.scores[pos][ctx]
return scores
def findMaskPositions(self, pos, modScores, noModScores):
maskPos = []
start = pos - self.post
end = pos + self.pre
for i in range(start, end + 1):
# Add a neighboring peak to the mask if
# a) it has a single-site qv > 20
# b) the observed IPDs are somewhat more likely under the modified
# hypothesis than the unmodified hypothesis
if i in self.rawKinetics and self.rawKinetics[i]["score"] > 20:
if modScores[i - start] - noModScores[i - start] > 1.0:
maskPos.append(i - pos)
return maskPos
def compareStates(self, current, prev):
return current[0:-1] == prev[1:]
|
