def dbranchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0.0)
dmodel1 = spidir.make_hky_deriv_matrix(bgfreq, kappa, t)
dmodel2 = spidir.make_hky_deriv_matrix(bgfreq, kappa, 0.0)
logl = 0.0
for j in xrange(seqlen):
ds = sum(
bgfreq[k]
* sum(dmodel1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
s = sum(
bgfreq[k]
* sum(model1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
logl += safediv(ds, s, INF)
return logl
def d2branchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0.0)
dmodel1 = spidir.make_hky_deriv_matrix(bgfreq, kappa, t)
dmodel2 = spidir.make_hky_deriv_matrix(bgfreq, kappa, 0.0)
d2model1 = spidir.make_hky_deriv2_matrix(bgfreq, kappa, t)
d2model2 = spidir.make_hky_deriv2_matrix(bgfreq, kappa, 0.0)
logl = 0.0
for j in xrange(seqlen):
g = sum(bgfreq[k] * sum(model1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
dg = sum(bgfreq[k] * sum(dmodel1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
d2g = sum(bgfreq[k] * sum(d2model1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
logl += - safediv(dg*dg, g*g, INF) + \
safediv(d2g, g, INF)
return logl
def branchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0)
logl = 0.0
for j in xrange(seqlen):
s = sum(bgfreq[k] * sum(model1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
logl += safelog(s, e)
return logl
def branchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0)
logl = 0.0
for j in xrange(seqlen):
s = sum(
bgfreq[k]
* sum(model1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
logl += safelog(s, e)
return logl
def test_hky(self):
"""general test"""
bgfreq = [.3, .2, .3, .2]
kappa = 2.0
t = 0.2
pprint(spidir.make_hky_matrix(bgfreq, kappa, t))
def test_hky(self):
"""general test"""
bgfreq = [.3, .2, .3, .2]
kappa = 2.0
t = 0.2
pprint(spidir.make_hky_matrix(bgfreq, kappa, t))
pprint(phylo.make_hky_matrix(t, bgfreq, kappa))
def test_hky_deriv(self):
"""general test"""
bgfreq = [.3, .2, .3, .2]
kappa = 2.0
i = random.randint(0, 3)
j = random.randint(0, 3)
x = list(frange(0, 1.0, .01))
y = [spidir.make_hky_matrix(bgfreq, kappa, t)[i][j] for t in x]
dy = [spidir.make_hky_deriv_matrix(bgfreq, kappa, t)[i][j] for t in x]
dy2 = [(spidir.make_hky_matrix(bgfreq, kappa, t + .01)[i][j] -
spidir.make_hky_matrix(bgfreq, kappa, t)[i][j]) / .01
for t in x]
prep_dir("test/output/hky")
rplot_start("test/output/hky/hky_deriv.pdf")
rplot("plot", x, y, t="l", ylim=[min(dy + y), max(dy + y)])
rp.lines(x, dy, col="red")
rp.lines(x, dy2, col="blue")
rplot_end(True)
def dbranchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0.0)
dmodel1 = spidir.make_hky_deriv_matrix(bgfreq, kappa, t)
dmodel2 = spidir.make_hky_deriv_matrix(bgfreq, kappa, 0.0)
logl = 0.0
for j in xrange(seqlen):
ds = sum(bgfreq[k] * sum(dmodel1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
s = sum(bgfreq[k] * sum(model1[k][x] * probs1[4 * j + x]
for x in xrange(4)) *
sum(model2[k][y] * probs2[4 * j + y]
for y in xrange(4)) for k in xrange(4))
logl += safediv(ds, s, INF)
return logl
def test_hky_deriv(self):
"""general test"""
bgfreq = [.3, .2, .3, .2]
kappa = 2.0
i = random.randint(0, 3)
j = random.randint(0, 3)
x = list(frange(0, 1.0, .01))
y = [spidir.make_hky_matrix(bgfreq, kappa, t)[i][j]
for t in x]
dy = [spidir.make_hky_deriv_matrix(bgfreq, kappa, t)[i][j]
for t in x]
dy2 = [(spidir.make_hky_matrix(bgfreq, kappa, t+.01)[i][j] -
spidir.make_hky_matrix(bgfreq, kappa, t)[i][j]) / .01
for t in x]
prep_dir("test/output/hky")
rplot_start("test/output/hky/hky_deriv.pdf")
rplot("plot", x, y, t="l", ylim=[min(dy + y), max(dy + y)])
rp.lines(x, dy, col="red")
rp.lines(x, dy2, col="blue")
rplot_end(True)
def d2branchlk(probs1, probs2, seqlen, bgfreq, kappa, t):
model1 = spidir.make_hky_matrix(bgfreq, kappa, t)
model2 = spidir.make_hky_matrix(bgfreq, kappa, 0.0)
dmodel1 = spidir.make_hky_deriv_matrix(bgfreq, kappa, t)
dmodel2 = spidir.make_hky_deriv_matrix(bgfreq, kappa, 0.0)
d2model1 = spidir.make_hky_deriv2_matrix(bgfreq, kappa, t)
d2model2 = spidir.make_hky_deriv2_matrix(bgfreq, kappa, 0.0)
logl = 0.0
for j in xrange(seqlen):
g = sum(
bgfreq[k]
* sum(model1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
dg = sum(
bgfreq[k]
* sum(dmodel1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
d2g = sum(
bgfreq[k]
* sum(d2model1[k][x] * probs1[4 * j + x] for x in xrange(4))
* sum(model2[k][y] * probs2[4 * j + y] for y in xrange(4))
for k in xrange(4)
)
logl += -safediv(dg * dg, g * g, INF) + safediv(d2g, g, INF)
return logl
def test_JC(self):
"""test equivalence to JC"""
bgfreq = [.25, .25, .25, .25]
kappa = 1.0
for t in frange(0, 1.0, .1):
mat = spidir.make_hky_matrix(bgfreq, kappa, t)
a = 1 / 3.0
r = (1 / 4.0) * (1 + 3 * exp(-4 * a * t))
s = (1 / 4.0) * (1 - exp(-4 * a * t))
mat2 = [[r, s, s, s], [s, r, s, s], [s, s, r, s], [s, s, s, r]]
for i in xrange(4):
for j in xrange(4):
fequal(mat[i][j], mat2[i][j])
def test1(self):
# (0, 1):2
# P(x_0,x_1|t) = \sum_{x_2} P(x_0|x_2,t_0) * P(x_1|x_2,t_1) * P(x_2)
# P(x_0|x_1,t) = P(x_0,x_1|t) / P(x_1|t) = P(x_0,x_1|t) / P(x_1)
for t in frange(0.1, 2.0, 0.1):
bgfreq = [0.25, 0.25, 0.25, 0.25]
kappa = 1.0
tree = treelib.parse_newick("(a:%f,b:%f)" % (t * 0.5, t * 0.5))
align = {"a": "C", "b": "T"}
l = spidir.calc_seq_likelihood_hky(tree, align, bgfreq, kappa) - log(0.25)
mat = spidir.make_hky_matrix(bgfreq, kappa, t)
l2 = log(mat[0][1])
print l, l2
fequal(l, l2)
def test1(self):
# (0, 1):2
# P(x_0,x_1|t) = \sum_{x_2} P(x_0|x_2,t_0) * P(x_1|x_2,t_1) * P(x_2)
# P(x_0|x_1,t) = P(x_0,x_1|t) / P(x_1|t) = P(x_0,x_1|t) / P(x_1)
for t in frange(.1, 2.0, .1):
bgfreq = [.25, .25, .25, .25]
kappa = 1.0
tree = treelib.parse_newick("(a:%f,b:%f)" % (t * .5, t * .5))
align = {"a": "C", "b": "T"}
l = spidir.calc_seq_likelihood_hky(tree, align, bgfreq, kappa) \
- log(.25)
mat = spidir.make_hky_matrix(bgfreq, kappa, t)
l2 = log(mat[0][1])
print l, l2
fequal(l, l2)
def test_jc(self):
"""test equivalence to JC"""
bgfreq = [.25, .25, .25, .25]
kappa = 1.0
for t in frange(0, 1.0, .1):
mat = spidir.make_hky_matrix(bgfreq, kappa, t)
a = 1/3.0
r = (1/4.0)*(1 + 3*exp(-4*a*t))
s = (1/4.0)*(1 - exp(-4*a*t))
mat2 = [[r, s, s, s],
[s, r, s, s],
[s, s, r, s],
[s, s, s, r]]
for i in xrange(4):
for j in xrange(4):
fequal(mat[i][j], mat2[i][j])
