-
Notifications
You must be signed in to change notification settings - Fork 40
/
MotifPssmPattern.py
225 lines (185 loc) · 7.6 KB
/
MotifPssmPattern.py
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
from Bio.Seq import Seq
from Bio import motifs
from Bio.Align.AlignInfo import PSSM
from .SequencePattern import SequencePattern
import numpy as np
class MotifPssmPattern(SequencePattern):
"""Motif pattern represented by a Position-Specific Scoring Matrix (PSSM).
This class is better instantiated by creating a pattern from a set of
sequence strings with ``MotifPssmPattern.from_sequences()``, or reading
pattern(s) for a file with ``MotifPssmPattern.from_file()``.
Parameters
----------
pssm
A Bio.Align.AlignInfo.PSSM object.
threshold
locations of the sequence with a PSSM score above this value will be
considered matches. For convenience, a relative_threshold can be
given instead.
relative_threshold
Value between 0 and 1 from which the threshold will be auto-computed.
0 means "match everything", 1 means "only match the one (or several)
sequence(s) with the absolute highest possible score".
"""
def __init__(
self, pssm, threshold=None, relative_threshold=None,
):
self.name = pssm.name
self.pssm = pssm.pssm
if relative_threshold is not None:
mini, maxi = self.pssm.min, self.pssm.max
threshold = mini + relative_threshold * (maxi - mini)
self.threshold = threshold
self.relative_threshold = relative_threshold
self.size = pssm.length
self.pssm_matrix = np.array([self.pssm[n] for n in "ATGC"])
self.is_palyndromic = False
@classmethod
def apply_pseudocounts(cls, motif, pseudocounts):
"""Add pseudocounts to the motif's pssm matrix.
Add nothing if pseudocounts is None, auto-compute pseudocounts if
pseudocounts="jaspar", else just attribute the pseudocounts as-is
to the motif.
"""
if pseudocounts is not None:
if pseudocounts == "jaspar":
pseudocounts = motifs.jaspar.calculate_pseudocounts(motif)
motif.pseudocounts = pseudocounts
def find_matches_in_string(self, sequence):
"""Find all positions where the PSSM score is above threshold."""
# NOTE: Before, I made my PSSM searches with Biopython. It was looong!
# Now I use Numpy and np.choice(), and I never looked back
# sequence = Seq(sequence, alphabet=alphabet)
# search = self.pssm.search(
# sequence, threshold=self.threshold, both=False
# )
indices = find_pssm_matches_with_numpy(
pssm_matrix=self.pssm_matrix, sequence=sequence, threshold=self.threshold,
)
return [(i, i + self.size, 1) for i in indices]
@classmethod
def from_sequences(
cls,
sequences,
name="unnamed",
pseudocounts="jaspar",
threshold=None,
relative_threshold=None,
):
"""Return a PSSM pattern computed from same-length sequences.
Parameters
----------
sequences
A list of same-length sequences.
name
Name to give to the pattern (will appear in reports etc.).
pseudocounts
Either a dict {"A": 0.01, "T": ...} or "jaspar" for automatic
pseudocounts from the Biopython.motifs.jaspar module (recommended),
or None for no pseudocounts at all (not recommended!).
threshold
locations of the sequence with a PSSM score above this value will be
considered matches. For convenience, a relative_threshold can be
given instead.
relative_threshold
Value between 0 and 1 from which the threshold will be auto-computed.
0 means "match everything", 1 means "only match the one (or several)
sequence(s) with the absolute highest possible score".
"""
sequences = [Seq(s) for s in sequences]
motif = motifs.create(sequences)
cls.apply_pseudocounts(motif, pseudocounts)
pssm = PSSM(motif.pssm)
pssm.name = name
return MotifPssmPattern(
pssm=pssm, threshold=threshold, relative_threshold=relative_threshold,
)
@classmethod
def list_from_file(
cls,
motifs_file,
file_format,
threshold=None,
pseudocounts="jaspar",
relative_threshold=None,
):
"""Return a list of PSSM patterns from a file in JASPAR, MEME, etc.
Parameters
----------
motifs_file
Path to a motifs file, or file handle.
file_format
File format. one of "jaspar", "meme", "TRANSFAC".
pseudocounts
Either a dict {"A": 0.01, "T": ...} or "jaspar" for automatic
pseudocounts from the Biopython.motifs.jaspar module (recommended),
or None for no pseudocounts at all (not recommended!).
threshold
locations of the sequence with a PSSM score above this value will be
considered matches. For convenience, a relative_threshold can be
given instead.
relative_threshold
Value between 0 and 1 from which the threshold will be auto-computed.
0 means "match everything", 1 means "only match the one (or several)
sequence(s) with the absolute highest possible score".
"""
if isinstance(motifs_file, str):
with open("./jaspar.txt", "r") as f:
motifs_list = motifs.parse(f, file_format)
else:
motifs_list = motifs.parse(motifs_file, file_format)
if pseudocounts is not None:
for motif in motifs_list:
cls.apply_pseudocounts(motif, pseudocounts)
return [
MotifPssmPattern(
pssm, threshold=threshold, relative_threshold=relative_threshold,
)
for pssm in motifs_list
]
def __str__(self):
if self.relative_threshold is not None:
threshold = "%d%%" % (100 * self.relative_threshold)
else:
threshold = "%.2f" % self.threshold
return "%s-PSSM(%s+)" % (self.name, threshold)
def __repr__(self):
if self.relative_threshold is not None:
threshold = "%d%%" % (100 * self.relative_threshold)
else:
threshold = "%.2f" % self.threshold
return "%s-PSSM(%s+)" % (self.name, threshold)
def find_pssm_matches_with_numpy(pssm_matrix, sequence, threshold):
"""Return every index in the +1 strand wit a PSSM score above threshold.
My numpy-based implementation is 10 times faster than Biopython for some
reason. Weird. Can someone else check?
Parameters:
-----------
pssm
A matrix whose rows give the frequency motif of ATGC (in this order).
sequence
A string representing a DNA sequence.
threshold
Every index with a score above this threshold will be returned.
"""
nucleotide_to_index = dict(zip("ATGC", range(4)))
len_pattern = len(pssm_matrix[0])
# If sequence is small, use normal python to avoid numpy overhead
if len(sequence) < 60:
nucl_indices = [nucleotide_to_index[n] for n in sequence]
return [
i
for i in range(len(sequence) - len_pattern)
if np.choose(nucl_indices[i : len_pattern + i], pssm_matrix).sum()
>= threshold
]
# If sequence is large, use Numpy for speed. tested experimentally
nucl_indices = np.array([nucleotide_to_index[n] for n in sequence], dtype="uint8")
len_pattern = len(pssm_matrix[0])
scores = np.array(
[
np.choose(nucl_indices[k : len_pattern + k], pssm_matrix).sum()
for k in range(len(sequence) - len_pattern)
]
)
return np.nonzero(scores >= threshold)[0]