/
ancient_genotypes_clustering.py
221 lines (209 loc) · 7.48 KB
/
ancient_genotypes_clustering.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
############################################################################
###########PARTITION########################################################
#CODE IS FROM http://jeromekelleher.net/generating-integer-partitions.html
def partition(n):
a = [0 for i in range(n + 1)]
k = 1
y = n - 1
while k != 0:
x = a[k - 1] + 1
k -= 1
while 2 * x <= y:
a[k] = x
y -= x
k += 1
l = k + 1
while x <= y:
a[k] = x
a[l] = y
yield a[:k + 2]
x += 1
y -= 1
a[k] = x + y
y = x + y - 1
yield a[:k + 1]
#CODE IS FROM http://stackoverflow.com/a/19369410
def slice_by_lengths(lengths, the_list):
for length in lengths:
new = []
for i in range(length):
new.append(the_list.pop(0))
yield new
def partitions(my_list):
partitions = partition(len(my_list))
permed = []
for each_partition in partitions:
permed.append(set(itertools.permutations(each_partition, len(each_partition))))
for each_tuple in itertools.chain(*permed):
yield list(slice_by_lengths(each_tuple, deepcopy(my_list)))
#########################################################################
#########################################################################
#This wasnts the "raw" reads data
def find_best_config(freq,reads,detail=False):
num_ind = len(reads)
pars = []
lnL = []
parts = []
for partition in partitions(range(num_ind)):
print "Processing %s"%partition
cur_opt = optimize_pop_params(freq,reads,partition,detail=detail)
cur_lnL = sum(map(lambda x: x[1],cur_opt))
cur_pars = map(lambda x: x[0], cur_opt)
parts.append(partition)
lnL.append(cur_lnL)
pars.append(cur_pars)
print 2*(2*len(partition))+cur_lnL
return parts, lnL, pars
#TODO: This does not guarantee that the number of clusters remains at k
#TODO: Probably need to implement a proper EM algorithm
#TODO: Proper EM might be hard.
#TODO: Maybe just hack so that if one cluster gets empty, you pop a dude out?
def cluster_anc(freq,reads,k,num_iter=10, detail=False):
num_ind = len(reads)
all_separate_pops = []
for i in range(num_ind):
all_separate_pops.append([i])
opts_separate = optimize_pop_params(freq,reads,all_separate_pops,detail=detail)
lnL_separate = map(lambda x: x[1],opts_separate)
freqs, reads_per_ind = make_read_dict_by_pop(freq,reads,all_separate_pops)
min_a, min_d, read_prob = bound_and_precompute_read_like(reads)
first_inds = rn.choice(num_ind,k)
params = [o[0] for o in np.array(opts_separate)[first_inds]]
pop_labels = np.zeros(num_ind)
for i in range(num_iter):
indLnLBest = []
for j in range(num_ind):
indLnL = np.full(k,-np.inf)
for l in range(k):
indLnL[l] = sum(compute_GT_like_DP(reads_per_ind[j],freqs, params[l][0],params[l][1],read_prob,min_a,min_d,detail=False))
pop_labels[j] = np.argmax(indLnL)
indLnLBest.append(-indLnL[pop_labels[j]])
new_pops = np.array([np.where(pop_labels==i)[0].tolist() for i in range(k)])
#this should make sure that every pop has a dude in it
for p,pop in enumerate(new_pops):
if np.array_equal(pop,[]):
new_guy = np.argmax(np.array(indLnLBest)-np.array(lnL_separate))
new_pops[pop_labels[new_guy]].remove(new_guy)
new_pops[p] = [new_guy]
indLnLBest[new_guy] = lnL_separate[new_guy]
#if i > 1 and new_pops == pops: break
pops = new_pops
print pop_labels, pops
opts = optimize_pop_params(freq,reads,pops,detail=detail)
params = [o[0] if o else None for o in opts]
return opts, pops
def chunk(num_ind,k):
seq = range(num_ind)
rn.shuffle(seq)
avg = num_ind/float(k)
out = []
last = 0.0
while last < num_ind:
out.append(sorted(seq[int(last):int(last+avg)]))
last += avg
return out
def cluster_k(freq, reads, k, num_iter=10, initialize = "random", detail=False):
num_ind = len(reads)
if initialize is "random":
first_inds = rn.choice(num_ind,k)
cur_pops = chunk(num_ind,k)
elif initialize is "kmeans":
all_separate = [[i] for i in range(num_ind)]
sep_opts = optimize_pop_params(freq,reads,all_separate,detail=detail)
pars = np.array(map(lambda x: x[0], sep_opts))
kmeans = cl.KMeans(n_clusters=k).fit(pars)
labels = kmeans.labels_
cur_pops = [[] for i in range(k)]
for i in range(len(labels)):
cur_pops[labels[i]].append(i)
else:
print "Unknown initialization procedure"
return 0
cur_opts = optimize_pop_params(freq,reads,cur_pops,detail=detail)
calculated = {}
for i in range(len(cur_pops)):
calculated[tuple(cur_pops[i])] = cur_opts[i]
print cur_pops, sum(map(lambda x: x[1], cur_opts))
for i in range(num_iter):
for ind in range(num_ind):
best_lamb = 0
best_pop = []
old_pop = map(lambda x: ind in x, cur_pops).index(True)
if len(cur_pops[old_pop]) == 1: continue
cur_minus = list(cur_pops[old_pop])
cur_minus.remove(ind)
if tuple(cur_minus) not in calculated:
cur_minus_opt = optimize_params_one_pop(freq,reads,cur_minus,detail=detail)
calculated[tuple(cur_minus)] = cur_minus_opt
else:
cur_minus_opt = calculated[tuple(cur_minus)]
for j in range(len(cur_pops)):
if ind in cur_pops[j]:
continue
cur_test = sorted([item for sublist in [cur_pops[j],[ind]] for item in sublist])
if tuple(cur_test) not in calculated:
cur_test_opt = optimize_params_one_pop(freq,reads,cur_test,detail=detail)
calculated[tuple(cur_test)] = cur_test_opt
else:
cur_test_opt = calculated[tuple(cur_test)]
new_lnL = cur_test_opt[1] + cur_minus_opt[1] #the one it's in now, the old one without it
old_lnL = cur_opts[j][1] + cur_opts[old_pop][1]
lamb = 2*(old_lnL-new_lnL)
print [cur_test, cur_minus], [cur_pops[j], cur_pops[old_pop]], lamb
if lamb > best_lamb:
best_lamb = lamb
best_pop = cur_test
best_test_opt = cur_test_opt
best_j = j
if best_lamb > 0:
cur_opts[best_j] = best_test_opt
cur_opts[old_pop] = cur_minus_opt
cur_pops[best_j] = best_pop
cur_pops[old_pop] = cur_minus
print cur_pops, sum(map(lambda x: x[1], cur_opts))
print cur_pops, sum(map(lambda x: x[1], cur_opts))
return cur_pops, cur_opts
def cluster_join(freq,reads,eps=1e-4,detail=False):
num_ind = len(reads)
cur_pops = []
for i in range(num_ind):
cur_pops.append([i])
cur_opts = optimize_pop_params(freq,reads,cur_pops,detail=detail-1)
any_to_merge = True
calculated = {}
while any_to_merge:
best_merge = []
best_lambda = 0
best_improv = 0
any_to_merge = False
for i in range(len(cur_pops)-1):
for j in range(i+1,len(cur_pops)):
cur_test = [item for sublist in [cur_pops[i],cur_pops[j]] for item in sublist]
print cur_test
if tuple(cur_test) not in calculated:
cur_test_opt = optimize_params_one_pop(freq,reads,cur_test,detail=detail-1)
calculated[tuple(cur_test)] = cur_test_opt
else:
cur_test_opt = calculated[tuple(cur_test)]
old_lnL = cur_opts[i][1] + cur_opts[j][1]
new_lnL = cur_test_opt[1]
new_lambda = 2*(old_lnL-new_lnL)
rel_improv = -(new_lnL/old_lnL-1)
if detail: print old_lnL, new_lnL, new_lambda, rel_improv
#if new_lambda > best_lambda:
if rel_improv > eps and rel_improv > best_improv:
best_merge = cur_test
best_i = i
best_j = j
best_opt = cur_test_opt
best_lambda = new_lambda
best_improv = rel_improv
any_to_merge = True
if best_lambda == 0: break
print best_merge, best_lambda, best_improv
cur_opts[best_i] = best_opt
cur_opts.pop(best_j)
cur_pops[best_i] = best_merge
cur_pops.pop(best_j)
print cur_pops
return cur_pops, cur_opts