def test_ml(self):
"""exercise the ML tree estimation"""
from numpy.testing import assert_allclose
aln = load_aligned_seqs(os.path.join(data_path, "brca1.fasta"), moltype="dna")
aln = aln.take_seqs(["Human", "Mouse", "Rat", "Dog"])
aln = aln.omit_gap_pos(allowed_gap_frac=0)
model = get_model("JC69")
lnL, tree = ML(model, aln).trex(a=3, k=1, show_progress=False)
assert_allclose(lnL, -8882.217502905267)
self.assertTrue(tree.same_topology(make_tree("(Mouse,Rat,(Human,Dog));")))
def TreeAlign(
model,
seqs,
tree=None,
indel_rate=0.01,
indel_length=0.01,
ui=None,
ests_from_pairwise=True,
param_vals=None,
):
"""Returns a multiple alignment and tree.
Uses the provided substitution model and a tree for determining the
progressive order. If a tree is not provided a Neighbour Joining tree is
constructed from pairwise distances estimated from pairwise aligning the
sequences. If running in parallel, only the distance estimation is
parallelised and only the master CPU returns the alignment and tree, other
CPU's return None, None.
Parameters
----------
model
a substitution model or the name of one, see available_models()
seqs
a sequence collection
indel_rate, indel_length
parameters for the progressive pair-HMM
ests_from_pairwise
if no tree provided and True, the median value
of the substitution model parameters are used
param_vals
named key, value pairs for model parameters. These
override ests_from_pairwise.
"""
from cogent3 import get_model
_exclude_params = ["mprobs", "rate", "bin_switch"]
param_vals = dict(param_vals) if param_vals else {}
seq_names = list(seqs.keys()) if isinstance(seqs, dict) else seqs.names
two_seqs = len(seq_names) == 2
model = get_model(model)
if tree:
tip_names = tree.get_tip_names()
tip_names.sort()
seq_names.sort()
assert (
tip_names == seq_names
), "names don't match between seqs and tree: tree=%s; seqs=%s" % (
tip_names,
seq_names,
)
ests_from_pairwise = False
elif two_seqs:
tree = make_tree(tip_names=seqs.names)
ests_from_pairwise = False
else:
if ests_from_pairwise:
est_params = [
param for param in model.get_param_list()
if param not in _exclude_params
]
else:
est_params = None
dcalc = EstimateDistances(seqs,
model,
do_pair_align=True,
est_params=est_params)
dcalc.run()
dists = dcalc.get_pairwise_distances().to_dict()
tree = NJ.nj(dists)
LF = model.make_likelihood_function(tree.bifurcating(name_unnamed=True),
aligned=False)
if ests_from_pairwise and not param_vals:
# we use the median to avoid the influence of outlier pairs
param_vals = {}
for param in est_params:
numbers = dcalc.get_param_values(param)
param_vals[param] = numbers.median
ui.display("Doing %s alignment" % ["progressive", "pairwise"][two_seqs])
with LF.updates_postponed():
for param, val in list(param_vals.items()):
LF.set_param_rule(param, value=val, is_constant=True)
LF.set_param_rule("indel_rate", value=indel_rate, is_constant=True)
LF.set_param_rule("indel_length", value=indel_length, is_constant=True)
LF.set_sequences(seqs)
lnL = LF.get_log_likelihood()
edge = lnL.edge
align = edge.get_viterbi_path().get_alignment()
align = align.to_moltype(model.moltype)
param_vals.update(
dict(
indel_length=indel_length,
indel_rate=indel_rate,
guide_tree=tree.get_newick(with_distances=True),
model=model.name,
lnL=lnL,
))
align.info["align_params"] = param_vals
return align, tree