| 1 |
import sys |
| 2 |
import csb.core |
| 3 |
import csb.bio.sequence as sequence |
| 4 |
|
| 5 |
from csb.bio.hmm import States, ScoreUnits, Transition, State |
| 6 |
|
| 7 |
|
| 8 |
GONNET = [10227, 3430, 2875, 3869, 1625, 2393, 4590, 6500, 2352, 3225, 5819, 4172, 1435, |
| 9 |
1579, 3728, 4610, 6264, 418, 1824, 5709, 3430, 7780, 2209, 2589, 584, 2369, |
| 10 |
3368, 3080, 2173, 1493, 3093, 5701, 763, 859, 1893, 2287, 3487, 444, 1338, 2356, |
| 11 |
2875, 2209, 3868, 3601, 501, 1541, 2956, 3325, 1951, 1065, 2012, 2879, 532, 688, |
| 12 |
1480, 2304, 3204, 219, 1148, 1759, 3869, 2589, 3601, 8618, 488, 2172, 6021, 4176, |
| 13 |
2184, 1139, 2151, 3616, 595, 670, 2086, 2828, 3843, 204, 1119, 2015, 1625, 584, |
| 14 |
501, 488, 5034, 355, 566, 900, 516, 741, 1336, 591, 337, 549, 419, 901, |
| 15 |
1197, 187, 664, 1373, 2393, 2369, 1541, 2172, 355, 1987, 2891, 1959, 1587, 1066, |
| 16 |
2260, 2751, 570, 628, 1415, 1595, 2323, 219, 871, 1682, 4590, 3368, 2956, 6021, |
| 17 |
566, 2891, 8201, 3758, 2418, 1624, 3140, 4704, 830, 852, 2418, 2923, 4159, 278, 1268, 2809, |
| 18 |
6500, 3080, 3325, 4176, 900, 1959, 3758, 26066, 2016, 1354, 2741, 3496, 741, 797, 2369, |
| 19 |
3863, 4169, 375, 1186, 2569, 2352, 2173, 1951, 2184, 516, 1587, 2418, 2016, 5409, |
| 20 |
1123, 2380, 2524, 600, 1259, 1298, 1642, 2446, 383, 876, 1691, 3225, 1493, 1065, |
| 21 |
1139, 741, 1066, 1624, 1354, 1123, 6417, 9630, 1858, 1975, 2225, 1260, 1558, 3131, |
| 22 |
417, 1697, 7504, 5819, 3093, 2012, 2151, 1336, 2260, 3140, 2741, 2380, 9630, 25113, |
| 23 |
3677, 4187, 5540, 2670, 2876, 5272, 1063, 3945, 11005, 4172, 5701, 2879, 3616, 591, |
| 24 |
2751, 4704, 3496, 2524, 1858, 3677, 7430, 949, 975, 2355, 2847, 4340, 333, 1451, 2932, |
| 25 |
1435, 763, 532, 595, 337, 570, 830, 741, 600, 1975, 4187, 949, 1300, 1111, 573, |
| 26 |
743, 1361, 218, 828, 2310, 1579, 859, 688, 670, 549, 628, 852, 797, 1259, 2225, |
| 27 |
5540, 975, 1111, 6126, 661, 856, 1498, 1000, 4464, 2602, 3728, 1893, 1480, 2086, 419, |
| 28 |
1415, 2418, 2369, 1298, 1260, 2670, 2355, 573, 661, 11834, 2320, 3300, 179, 876, 2179, |
| 29 |
4610, 2287, 2304, 2828, 901, 1595, 2923, 3863, 1642, 1558, 2876, 2847, 743, 856, 2320, |
| 30 |
3611, 4686, 272, 1188, 2695, 6264, 3487, 3204, 3843, 1197, 2323, 4159, 4169, 2446, 3131, |
| 31 |
5272, 4340, 1361, 1498, 3300, 4686, 8995, 397, 1812, 5172, 418, 444, 219, 204, 187, |
| 32 |
219, 278, 375, 383, 417, 1063, 333, 218, 1000, 179, 272, 397, 4101, 1266, 499, |
| 33 |
1824, 1338, 1148, 1119, 664, 871, 1268, 1186, 876, 1697, 3945, 1451, 828, 4464, 876, |
| 34 |
1188, 1812, 1266, 9380, 2227, 5709, 2356, 1759, 2015, 1373, 1682, 2809, 2569, 1691, 7504, |
| 35 |
11005, 2932, 2310, 2602, 2179, 2695, 5172, 499, 2227.0, 11569.0] |
| 36 |
""" |
| 37 |
Gonnet matrix frequencies taken from HHpred |
| 38 |
""" |
| 39 |
|
| 40 |
|
| 41 |
class PseudocountBuilder(object): |
| 42 |
""" |
| 43 |
Constructs profile HMMs with pseudocounts. |
| 44 |
""" |
| 45 |
|
| 46 |
def __init__(self, hmm): |
| 47 |
self._hmm = hmm |
| 48 |
|
| 49 |
@property |
| 50 |
def hmm(self): |
| 51 |
return self._hmm |
| 52 |
|
| 53 |
def add_emission_pseudocounts(self, tau=0.1, pca=2.5, pcb=0.5, pcc=1.0): |
| 54 |
""" |
| 55 |
Port from HHpred, it uses the conditional background probabilities, |
| 56 |
inferred from the Gonnet matrix. |
| 57 |
|
| 58 |
@param tau: admission weight, i.e how much of the final score is |
| 59 |
determined by the background probabilities. |
| 60 |
0.0=no pseudocounts. |
| 61 |
@type tau: float |
| 62 |
""" |
| 63 |
from numpy import array, dot, transpose, clip |
| 64 |
|
| 65 |
if self.hmm.pseudocounts or self.hmm.emission_pseudocounts: |
| 66 |
return |
| 67 |
if abs(tau) < 1e-6: |
| 68 |
return |
| 69 |
|
| 70 |
# Assume probabilities |
| 71 |
if not self.hmm.score_units == ScoreUnits.Probability: |
| 72 |
self.hmm.convert_scores(units=ScoreUnits.Probability) |
| 73 |
|
| 74 |
alphabet = csb.core.Enum.values(sequence.StdProteinAlphabet) |
| 75 |
|
| 76 |
## S = SubstitutionMatrix(substitution_matrix) |
| 77 |
s_mat = array(GONNET) |
| 78 |
#Normalize |
| 79 |
s_mat /= s_mat.sum() |
| 80 |
s_mat = s_mat.reshape((len(alphabet), len(alphabet))) |
| 81 |
# Marginalize matrix |
| 82 |
s_marginal = s_mat.sum(-1) |
| 83 |
s_conditional = s_mat / s_marginal |
| 84 |
# Get data and info from hmm |
| 85 |
em = array([ [layer[States.Match].emission[aa] or 0.0 for aa in alphabet] |
| 86 |
for layer in self.hmm.layers]) |
| 87 |
|
| 88 |
em = clip(em, sys.float_info.min, 1.) |
| 89 |
|
| 90 |
neff_m = array([l.effective_matches for l in self.hmm.layers]) |
| 91 |
|
| 92 |
g = dot(em, transpose(s_conditional)) |
| 93 |
|
| 94 |
if neff_m is not None: |
| 95 |
tau = clip(pca / (1. + (neff_m / pcb) ** pcc), 0.0, pcc) |
| 96 |
e = transpose((1. - tau) * transpose(em) + tau * transpose(g)) |
| 97 |
else: |
| 98 |
e = (1. - tau) * em + tau * g |
| 99 |
|
| 100 |
# Renormalize e |
| 101 |
e = transpose(transpose(e) / e.sum(-1)) |
| 102 |
|
| 103 |
for i, layer in enumerate(self.hmm.layers): |
| 104 |
layer[States.Match].emission.set(dict(zip(alphabet, e[i]))) |
| 105 |
|
| 106 |
self.hmm.emission_pseudocounts = True |
| 107 |
return |
| 108 |
|
| 109 |
|
| 110 |
|
| 111 |
def add_transition_pseudocounts(self, gapb=1., gapd=0.15, gape=1.0, gapf=0.6, gapg=0.6, gapi=0.6): |
| 112 |
""" |
| 113 |
Add pseudocounts to the transitions. A port from hhsearch |
| 114 |
-gapb 1.0 -gapd 0.15 -gape 1.0 -gapf 0.6 -gapg 0.6 -gapi 0.6 |
| 115 |
""" |
| 116 |
|
| 117 |
from numpy import array |
| 118 |
|
| 119 |
if not self.hmm._score_units == ScoreUnits.Probability: |
| 120 |
self.hmm.convert_scores(units=ScoreUnits.Probability) |
| 121 |
|
| 122 |
if self.hmm.pseudocounts or self.hmm.transition_pseudocounts: |
| 123 |
return |
| 124 |
|
| 125 |
# We need a fully populated HMM so first add all missing states |
| 126 |
states = [States.Match, States.Insertion, States.Deletion] |
| 127 |
background = self.hmm.layers[1][States.Match].background |
| 128 |
for layer in self.hmm.layers: |
| 129 |
rank = layer.rank |
| 130 |
for state in states: |
| 131 |
if state not in layer: |
| 132 |
|
| 133 |
if state is States.Deletion: |
| 134 |
# Add a new Deletion state |
| 135 |
deletion = State(States.Deletion) |
| 136 |
deletion.rank = rank |
| 137 |
layer.append(deletion) |
| 138 |
|
| 139 |
elif state is States.Insertion: |
| 140 |
# Add a new Deletion state |
| 141 |
insertion = State(States.Insertion, |
| 142 |
emit=csb.core.Enum.members( |
| 143 |
sequence.SequenceAlphabets.Protein)) |
| 144 |
insertion.background.set(background) |
| 145 |
insertion.emission.set(background) |
| 146 |
insertion.rank = rank |
| 147 |
layer.append(insertion) |
| 148 |
|
| 149 |
if not self.hmm.start_insertion: |
| 150 |
insertion = State(States.Insertion, |
| 151 |
emit=csb.core.Enum.members( |
| 152 |
sequence.SequenceAlphabets.Protein)) |
| 153 |
insertion.background.set(background) |
| 154 |
insertion.emission.set(background) |
| 155 |
insertion.rank = 0 |
| 156 |
self.hmm.start_insertion = insertion |
| 157 |
|
| 158 |
# make hmm completly connected |
| 159 |
for i in range(1, self.hmm.layers.length): |
| 160 |
layer = self.hmm.layers[i] |
| 161 |
#Start with match state |
| 162 |
state = layer[States.Match] |
| 163 |
if not States.Insertion in state.transitions: |
| 164 |
state.transitions.append(Transition(state, |
| 165 |
self.hmm.layers[i][States.Insertion], |
| 166 |
0.0)) |
| 167 |
if not States.Deletion in state.transitions: |
| 168 |
state.transitions.append(Transition(state, |
| 169 |
self.hmm.layers[i + 1][States.Deletion], |
| 170 |
0.0)) |
| 171 |
state = layer[States.Insertion] |
| 172 |
if not States.Insertion in state.transitions: |
| 173 |
state.transitions.append(Transition(state, |
| 174 |
self.hmm.layers[i][States.Insertion], |
| 175 |
0.0)) |
| 176 |
if not States.Match in state.transitions: |
| 177 |
state.transitions.append(Transition(state, |
| 178 |
self.hmm.layers[i + 1][States.Match], |
| 179 |
0.0)) |
| 180 |
state = layer[States.Deletion] |
| 181 |
if not States.Deletion in state.transitions: |
| 182 |
state.transitions.append(Transition(state, |
| 183 |
self.hmm.layers[i + 1][States.Deletion], |
| 184 |
0.0)) |
| 185 |
if not States.Match in state.transitions: |
| 186 |
state.transitions.append(Transition(state, |
| 187 |
self.hmm.layers[i + 1][States.Match], |
| 188 |
0.0)) |
| 189 |
# start layer |
| 190 |
state = self.hmm.start |
| 191 |
if not States.Insertion in self.hmm.start.transitions: |
| 192 |
state.transitions.append(Transition(self.hmm.start, |
| 193 |
self.hmm.start_insertion, |
| 194 |
0.0)) |
| 195 |
if not States.Deletion in self.hmm.start.transitions: |
| 196 |
state.transitions.append(Transition(self.hmm.start, |
| 197 |
self.hmm.layers[1][States.Deletion], |
| 198 |
0.0)) |
| 199 |
|
| 200 |
state = self.hmm.start_insertion |
| 201 |
if not States.Insertion in self.hmm.start_insertion.transitions: |
| 202 |
state.transitions.append(Transition(self.hmm.start_insertion, |
| 203 |
self.hmm.start_insertion, |
| 204 |
0.0)) |
| 205 |
if not States.Match in self.hmm.start_insertion.transitions: |
| 206 |
state.transitions.append(Transition(self.hmm.start_insertion, |
| 207 |
self.hmm.layers[1][States.Match], |
| 208 |
0.0)) |
| 209 |
|
| 210 |
# last layer |
| 211 |
state = self.hmm.layers[-1][States.Match] |
| 212 |
if not States.Insertion in state.transitions: |
| 213 |
state.transitions.append(Transition(state, |
| 214 |
self.hmm.layers[-1][States.Insertion], |
| 215 |
0.0)) |
| 216 |
state = self.hmm.layers[-1][States.Insertion] |
| 217 |
if not States.Insertion in state.transitions: |
| 218 |
state.transitions.append(Transition(state, |
| 219 |
self.hmm.layers[-1][States.Insertion], |
| 220 |
0.0)) |
| 221 |
|
| 222 |
if not States.End in state.transitions: |
| 223 |
state.transitions.append(Transition(state, |
| 224 |
self.hmm.end, |
| 225 |
0.0)) |
| 226 |
state = self.hmm.layers[-1][States.Deletion] |
| 227 |
if not States.End in state.transitions: |
| 228 |
state.transitions.append(Transition(state, |
| 229 |
self.hmm.end, |
| 230 |
0.0)) |
| 231 |
|
| 232 |
|
| 233 |
|
| 234 |
# Now we have created a fully connected HMM |
| 235 |
# Lates add pseuod counts |
| 236 |
# Calculate pseudo counts |
| 237 |
|
| 238 |
# to be honest I really do not know how they came up with this |
| 239 |
pc_MD = pc_MI = 0.0286 * gapd |
| 240 |
pc_MM = 1. - 2 * pc_MD |
| 241 |
pc_DD = pc_II = gape / (gape - 1 + 1 / 0.75) |
| 242 |
pc_DM = pc_IM = 1. - pc_II |
| 243 |
|
| 244 |
|
| 245 |
# Get current transtion probabilities |
| 246 |
t_mm = self.hmm.start.transitions[States.Match].probability |
| 247 |
t_mi = self.hmm.start.transitions[States.Insertion].probability |
| 248 |
t_md = self.hmm.start.transitions[States.Deletion].probability |
| 249 |
|
| 250 |
# Transitions from Match state |
| 251 |
n_eff = self.hmm.effective_matches |
| 252 |
|
| 253 |
t = array([(n_eff - 1) * t_mm + gapb * pc_MM, |
| 254 |
(n_eff - 1) * t_mi + gapb * pc_MI, |
| 255 |
(n_eff - 1) * t_md + gapb * pc_MD]) |
| 256 |
# normalize to one |
| 257 |
t /= t.sum() |
| 258 |
# Set |
| 259 |
self.hmm.start.transitions[States.Match].probability = t[0] |
| 260 |
self.hmm.start.transitions[States.Insertion].probability = t[1] |
| 261 |
self.hmm.start.transitions[States.Deletion].probability = t[2] |
| 262 |
|
| 263 |
# Rinse and repeat |
| 264 |
t_im = self.hmm.start_insertion.transitions[States.Match].probability |
| 265 |
t_ii = self.hmm.start_insertion.transitions[States.Insertion].probability |
| 266 |
|
| 267 |
t = array([t_im + gapb * pc_IM, t_ii + gapb * pc_II]) |
| 268 |
t /= t.sum() |
| 269 |
|
| 270 |
self.hmm.start_insertion.transitions[States.Match].probability = t[0] |
| 271 |
t_ii = self.hmm.start_insertion.transitions[States.Insertion].probability = t[1] |
| 272 |
|
| 273 |
# And now for all layers |
| 274 |
for layer in self.hmm.layers[:-1]: |
| 275 |
# Get current transtion probabilities |
| 276 |
t_mm = layer[States.Match].transitions[States.Match].probability |
| 277 |
t_mi = layer[States.Match].transitions[States.Insertion].probability |
| 278 |
t_md = layer[States.Match].transitions[States.Deletion].probability |
| 279 |
n_eff = layer.effective_matches |
| 280 |
t = array([(n_eff - 1) * t_mm + gapb * pc_MM, |
| 281 |
(n_eff - 1) * t_mi + gapb * pc_MI, |
| 282 |
(n_eff - 1) * t_md + gapb * pc_MD]) |
| 283 |
# normalize to one |
| 284 |
t /= t.sum() |
| 285 |
layer[States.Match].transitions[States.Match].probability = t[0] |
| 286 |
layer[States.Match].transitions[States.Insertion].probability = t[1] |
| 287 |
layer[States.Match].transitions[States.Deletion].probability = t[2] |
| 288 |
|
| 289 |
# Transitions from insert state |
| 290 |
t_im = layer[States.Insertion].transitions[States.Match].probability |
| 291 |
t_ii = layer[States.Insertion].transitions[States.Insertion].probability |
| 292 |
n_eff = layer.effective_insertions |
| 293 |
t = array([t_im * n_eff + gapb * pc_IM, |
| 294 |
t_im * n_eff + gapb * pc_II]) |
| 295 |
# normalize to one |
| 296 |
t /= t.sum() |
| 297 |
layer[States.Insertion].transitions[States.Match].probability = t[0] |
| 298 |
layer[States.Insertion].transitions[States.Insertion].probability = t[1] |
| 299 |
|
| 300 |
# Transitions form deletion state |
| 301 |
t_dm = layer[States.Deletion].transitions[States.Match].probability |
| 302 |
t_dd = layer[States.Deletion].transitions[States.Deletion].probability |
| 303 |
n_eff = layer.effective_deletions |
| 304 |
t = array([t_dm * n_eff + gapb * pc_DM, |
| 305 |
t_dd * n_eff + gapb * pc_DD]) |
| 306 |
# normalize to one |
| 307 |
t /= t.sum() |
| 308 |
layer[States.Deletion].transitions[States.Match].probability = t[0] |
| 309 |
layer[States.Deletion].transitions[States.Deletion].probability = t[1] |
| 310 |
|
| 311 |
#Last layer |
| 312 |
|
| 313 |
layer = self.hmm.layers[-1] |
| 314 |
t_mm = layer[States.Match].transitions[States.End].probability |
| 315 |
t_mi = layer[States.Match].transitions[States.Insertion].probability |
| 316 |
n_eff = layer.effective_matches |
| 317 |
# No deletion |
| 318 |
t = array([(n_eff - 1) * t_mm + gapb * pc_MM, |
| 319 |
(n_eff - 1) * t_mi + gapb * pc_MI]) |
| 320 |
# normalize to one |
| 321 |
t /= t.sum() |
| 322 |
layer[States.Match].transitions[States.End].probability = t[0] |
| 323 |
layer[States.Match].transitions[States.Insertion].probability = t[1] |
| 324 |
|
| 325 |
# Transitions from insert state |
| 326 |
t_im = layer[States.Insertion].transitions[States.End].probability |
| 327 |
t_ii = layer[States.Insertion].transitions[States.Insertion].probability |
| 328 |
n_eff = layer.effective_insertions |
| 329 |
t = array([t_im * n_eff + gapb * pc_IM, |
| 330 |
t_im * n_eff + gapb * pc_II]) |
| 331 |
# normalize to one |
| 332 |
t /= t.sum() |
| 333 |
layer[States.Insertion].transitions[States.End].probability = t[0] |
| 334 |
layer[States.Insertion].transitions[States.Insertion].probability = t[1] |
| 335 |
|
| 336 |
layer[States.Deletion].transitions[States.End].probability = 1. |
| 337 |
|
| 338 |
self.hmm.transition_pseudocounts = True |
| 339 |
return |
| 340 |
|