def test_prior(self):
"""
Calculate state priors
"""
k = 10
n = 1e4
rho = 1.5e-8
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points()
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
prior = [model.prob_prior(0, j)
for j in xrange(model.get_num_states(0))]
print prior
print sum(map(exp, prior))
fequal(sum(map(exp, prior)), 1.0, rel=.01)
def test_trans_single(self):
"""
Calculate transition probabilities
Only calculate a single matrix
"""
k = 4
n = 1e4
rho = 1.5e-8
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
print "recomb", arglib.get_recomb_pos(arg)
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
pos = 10
tree = arg.get_marginal_tree(pos)
mat = arghmm.calc_transition_probs(tree, model.states[pos],
model.nlineages, model.times,
model.time_steps, model.popsizes,
rho)
print model.states[pos]
pc(mat)
for row in mat:
print sum(map(exp, row))
def test_emit(self):
"""
Calculate emission probabilities
"""
k = 10
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8 * 20
length = int(1e3) / 20
times = arghmm.get_time_points(ntimes=20, maxtime=200000)
arg = arghmm.sample_arg_dsmc(k, 2*n, rho, start=0, end=length,
times=times)
muts = arghmm.sample_arg_mutations(arg, mu, times)
seqs = arghmm.make_alignment(arg, muts)
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
trees, names = arghmm.arg2ctrees(arg, times)
seqs2, nseqs, seqlen = arghmm.seqs2cseqs(seqs, names + [new_name])
assert arghmm.arghmm_assert_emit(trees, len(times), times, mu,
seqs2, nseqs, seqlen)
def test_emit(self):
"""
Calculate emission probabilities
"""
k = 10
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8 * 20
length = int(1e3) / 20
times = arghmm.get_time_points(ntimes=20, maxtime=200000)
arg = arghmm.sample_arg_dsmc(k,
2 * n,
rho,
start=0,
end=length,
times=times)
muts = arghmm.sample_arg_mutations(arg, mu, times)
seqs = arghmm.make_alignment(arg, muts)
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
trees, names = arghmm.arg2ctrees(arg, times)
seqs2, nseqs, seqlen = arghmm.seqs2cseqs(seqs, names + [new_name])
assert arghmm.arghmm_assert_emit(trees, len(times), times, mu, seqs2,
nseqs, seqlen)
def test_trans_single(self):
"""
Calculate transition probabilities
Only calculate a single matrix
"""
k = 4
n = 1e4
rho = 1.5e-8
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
print "recomb", arglib.get_recomb_pos(arg)
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
pos = 10
tree = arg.get_marginal_tree(pos)
mat = arghmm.calc_transition_probs(
tree, model.states[pos], model.nlineages,
model.times, model.time_steps, model.popsizes, rho)
print model.states[pos]
pc(mat)
for row in mat:
print sum(map(exp, row))
def test_prior(self):
"""
Calculate state priors
"""
k = 10
n = 1e4
rho = 1.5e-8
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points()
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
prior = [
model.prob_prior(0, j) for j in xrange(model.get_num_states(0))
]
print prior
print sum(map(exp, prior))
fequal(sum(map(exp, prior)), 1.0, rel=.01)
def test_post_plot(self):
k = 6
n = 1e4
rho = 1.5e-8 * 50
mu = 2.5e-8 * 50
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=30)
arghmm.discretize_arg(arg, times)
pause()
# save
#arglib.write_arg("test/data/k4.arg", arg)
#fasta.write_fasta("test/data/k4.fa", seqs)
new_name = "n%d" % (k - 1)
thread = list(
arghmm.iter_chrom_thread(arg, arg[new_name], by_block=False))
p = plot(cget(thread, 1), style="lines", ymin=times[1], ylog=10)
# remove chrom
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg,
seqs,
new_name=new_name,
times=times,
rho=rho,
mu=mu)
print "states", len(model.states[0])
print "muts", len(muts)
print "recomb", len(model.recomb_pos) - 2, model.recomb_pos[1:-1]
p.plot(model.recomb_pos, [10000] * len(model.recomb_pos),
style="points")
probs = arghmm.get_posterior_probs(model, length, verbose=True)
print "done"
high = list(arghmm.iter_posterior_times(model, probs, .95))
low = list(arghmm.iter_posterior_times(model, probs, .05))
p.gnuplot("set linestyle 2")
p.plot(high, style="lines")
p.gnuplot("set linestyle 2")
p.plot(low, style="lines")
#write_list("test/data/post_real.txt", cget(thread, 1))
#write_list("test/data/post_high.txt", high)
#write_list("test/data/post_low.txt", low)
pause()
def test_post_plot(self):
k = 6
n = 1e4
rho = 1.5e-8 * 50
mu = 2.5e-8 * 50
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=30)
arghmm.discretize_arg(arg, times)
pause()
# save
#arglib.write_arg("test/data/k4.arg", arg)
#fasta.write_fasta("test/data/k4.fa", seqs)
new_name = "n%d" % (k-1)
thread = list(arghmm.iter_chrom_thread(arg, arg[new_name],
by_block=False))
p = plot(cget(thread, 1), style="lines", ymin=times[1],
ylog=10)
# remove chrom
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times,
rho=rho, mu=mu)
print "states", len(model.states[0])
print "muts", len(muts)
print "recomb", len(model.recomb_pos) - 2, model.recomb_pos[1:-1]
p.plot(model.recomb_pos, [10000] * len(model.recomb_pos),
style="points")
probs = arghmm.get_posterior_probs(model, length, verbose=True)
print "done"
high = list(arghmm.iter_posterior_times(model, probs, .95))
low = list(arghmm.iter_posterior_times(model, probs, .05))
p.gnuplot("set linestyle 2")
p.plot(high, style="lines")
p.gnuplot("set linestyle 2")
p.plot(low, style="lines")
#write_list("test/data/post_real.txt", cget(thread, 1))
#write_list("test/data/post_high.txt", high)
#write_list("test/data/post_low.txt", low)
pause()
def test_trans_switch_single(self):
"""
Calculate transitions probabilities for switching between blocks
Only calculate a single matrix
"""
k = 5
n = 1e4
rho = 1.5e-8 * 100
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
#arglib.write_arg("tmp/a.arg", arg)
#arg = arglib.read_arg("tmp/a.arg")
#arg.set_ancestral()
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(5)
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
# get recombs
recombs = list(x.pos for x in arghmm.iter_visible_recombs(arg))
print "recomb", recombs
pos = recombs[0] + 1
tree = arg.get_marginal_tree(pos - .5)
last_tree = arg.get_marginal_tree(pos - 1 - .5)
print "states1>>", model.states[pos - 1]
print "states2>>", model.states[pos]
treelib.draw_tree_names(last_tree.get_tree(), minlen=5, maxlen=5)
treelib.draw_tree_names(tree.get_tree(), minlen=5, maxlen=5)
print "pos>>", pos
recomb = [x for x in tree
if x.event == "recomb" and x.pos + 1 == pos][0]
mat = arghmm.calc_transition_probs_switch(tree, last_tree, recomb.name,
model.states[pos - 1],
model.states[pos],
model.nlineages, model.times,
model.time_steps,
model.popsizes, rho)
pc(mat)
def test_trans_switch_single(self):
"""
Calculate transitions probabilities for switching between blocks
Only calculate a single matrix
"""
k = 5
n = 1e4
rho = 1.5e-8 * 100
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
#arglib.write_arg("tmp/a.arg", arg)
#arg = arglib.read_arg("tmp/a.arg")
#arg.set_ancestral()
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(5)
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
# get recombs
recombs = list(x.pos for x in arghmm.iter_visible_recombs(arg))
print "recomb", recombs
pos = recombs[0] + 1
tree = arg.get_marginal_tree(pos-.5)
last_tree = arg.get_marginal_tree(pos-1-.5)
print "states1>>", model.states[pos-1]
print "states2>>", model.states[pos]
treelib.draw_tree_names(last_tree.get_tree(), minlen=5, maxlen=5)
treelib.draw_tree_names(tree.get_tree(), minlen=5, maxlen=5)
print "pos>>", pos
recomb = [x for x in tree
if x.event == "recomb" and x.pos+1 == pos][0]
mat = arghmm.calc_transition_probs_switch(
tree, last_tree, recomb.name,
model.states[pos-1], model.states[pos],
model.nlineages, model.times,
model.time_steps, model.popsizes, rho)
pc(mat)
def test_state_corr(self):
k = 12
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8 * 20
length = int(1e3) / 20
times = arghmm.get_time_points(ntimes=20, maxtime=200e3)
arg = arghmm.sample_arg_dsmc(k,
2 * n,
rho,
start=0,
end=length,
times=times)
muts = arghmm.sample_arg_mutations(arg, mu, times)
seqs = arglib.make_alignment(arg, muts)
# remove chrom
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg,
seqs,
new_name=new_name,
times=times,
rho=rho,
mu=mu)
print "states", len(model.states[0])
nstates = len(model.states[0])
prior = [-util.INF] * nstates
prior[random.randint(0, nstates)] = 0.0
probs1 = list(arghmm.forward_algorithm(model, length, verbose=True))
probs2 = list(
arghmm.forward_algorithm(model, length, prior=prior, verbose=True))
model.rho *= 1e-9
probs3 = list(
arghmm.forward_algorithm(model, length, prior=prior, verbose=True))
p = plot(vsubs(probs1[length - 1], mean(probs1[length - 1])))
p.plot(vsubs(probs2[length - 1], mean(probs2[length - 1])))
p.plot(vsubs(probs3[length - 1], mean(probs3[length - 1])))
pause()
def test_backward(self):
"""
Run backward algorithm
"""
k = 3
n = 1e4
rho = 1.5e-8 * 100
mu = 2.5e-8 * 100
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=10)
arghmm.discretize_arg(arg, times)
tree = arg.get_marginal_tree(0)
print tree.root.age
treelib.draw_tree_names(tree.get_tree(), minlen=5, scale=4e-4)
# remove chrom
new_name = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg,
seqs,
new_name=new_name,
times=times,
rho=rho,
mu=mu)
print "states", len(model.states[0])
print "recomb", model.recomb_pos
print "muts", len(muts)
probs = hmm.backward_algorithm(model, length, verbose=True)
for pcol in probs:
p = sum(map(exp, pcol))
print p, " ".join("%.3f" % f for f in map(exp, pcol))
def test_state_corr(self):
k = 12
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8 * 20
length = int(1e3) / 20
times = arghmm.get_time_points(ntimes=20, maxtime=200e3)
arg = arghmm.sample_arg_dsmc(k, 2*n, rho, start=0, end=length,
times=times)
muts = arghmm.sample_arg_mutations(arg, mu, times)
seqs = arglib.make_alignment(arg, muts)
# remove chrom
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times,
rho=rho, mu=mu)
print "states", len(model.states[0])
nstates = len(model.states[0])
prior = [-util.INF] * nstates
prior[random.randint(0, nstates)] = 0.0
probs1 = list(arghmm.forward_algorithm(model, length, verbose=True))
probs2 = list(arghmm.forward_algorithm(model, length, prior=prior,
verbose=True))
model.rho *= 1e-9
probs3 = list(arghmm.forward_algorithm(model, length, prior=prior,
verbose=True))
p = plot(vsubs(probs1[length-1], mean(probs1[length-1])))
p.plot(vsubs(probs2[length-1], mean(probs2[length-1])))
p.plot(vsubs(probs3[length-1], mean(probs3[length-1])))
pause()
def test_emit_argmax(self):
"""
Calculate emission probabilities
"""
k = 10
n = 1e4
rho = 0.0
mu = 2.5e-8 * 100
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k - 1)
thread = list(arghmm.iter_chrom_thread(arg, arg[new_name]))
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
nstates = model.get_num_states(1)
probs = [0.0 for j in xrange(nstates)]
for i in xrange(1, length):
if i % 100 == 0:
print i
for j in xrange(nstates):
probs[j] += model.prob_emission(i, j)
print
# is the maximum likelihood emission matching truth
data = sorted(zip(probs, model.states[0]), reverse=True)
pc(data[:20])
state = (thread[0][0], times.index(thread[0][1]))
print data[0][1], state
assert data[0][1] == state
def test_emit_argmax(self):
"""
Calculate emission probabilities
"""
k = 10
n = 1e4
rho = 0.0
mu = 2.5e-8 * 100
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
new_name = "n%d" % (k-1)
thread = list(arghmm.iter_chrom_thread(arg, arg[new_name]))
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times)
nstates = model.get_num_states(1)
probs = [0.0 for j in xrange(nstates)]
for i in xrange(1, length):
if i % 100 == 0:
print i
for j in xrange(nstates):
probs[j] += model.prob_emission(i, j)
print
# is the maximum likelihood emission matching truth
data = sorted(zip(probs, model.states[0]), reverse=True)
pc(data[:20])
state = (thread[0][0], times.index(thread[0][1]))
print data[0][1], state
assert data[0][1] == state
def test_remove_thread(self):
"""
Remove a leaf from an ARG
"""
k = 3
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
chrom = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, chrom)
recomb = arglib.get_recomb_pos(arg)
print "recomb", recomb
tree = arg.get_marginal_tree(-.5)
draw_tree_names(tree.get_tree(), minlen=5, maxlen=5)
print sorted([(x.pos, x.event) for x in tree])
def test_remove_thread(self):
"""
Remove a leaf from an ARG
"""
k = 3
n = 1e4
rho = 1.5e-8 * 20
mu = 2.5e-8
length = 1000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
times = arghmm.get_time_points(10)
arghmm.discretize_arg(arg, times)
chrom = "n%d" % (k - 1)
arg = arghmm.remove_arg_thread(arg, chrom)
recomb = arglib.get_recomb_pos(arg)
print "recomb", recomb
tree = arg.get_marginal_tree(-.5)
draw_tree_names(tree.get_tree(), minlen=5, maxlen=5)
print sorted([(x.pos, x.event) for x in tree])
def test_backward(self):
"""
Run backward algorithm
"""
k = 3
n = 1e4
rho = 1.5e-8 * 100
mu = 2.5e-8 * 100
length = 10000
arg = arglib.sample_arg(k, n, rho, start=0, end=length)
muts = arglib.sample_arg_mutations(arg, mu)
seqs = arglib.make_alignment(arg, muts)
times = arghmm.get_time_points(ntimes=10)
arghmm.discretize_arg(arg, times)
tree = arg.get_marginal_tree(0)
print tree.root.age
treelib.draw_tree_names(tree.get_tree(), minlen=5, scale=4e-4)
# remove chrom
new_name = "n%d" % (k-1)
arg = arghmm.remove_arg_thread(arg, new_name)
model = arghmm.ArgHmm(arg, seqs, new_name=new_name, times=times,
rho=rho, mu=mu)
print "states", len(model.states[0])
print "recomb", model.recomb_pos
print "muts", len(muts)
probs = hmm.backward_algorithm(model, length, verbose=True)
for pcol in probs:
p = sum(map(exp, pcol))
print p, " ".join("%.3f" % f for f in map(exp, pcol))
