def run_method(method, tree, m = 300, kappa = 2, mutation_rate=0.05, threshold = None, verbose = False):
start_time = time.time()
observations, taxa_meta = generation.simulate_sequences(m, tree_model=tree, seq_model=generation.HKY(kappa = kappa), mutation_rate=mutation_rate, alphabet="DNA")
runtime = time.time() - start_time
print("Simulation took %s seconds" % runtime)
if method == "RaXML":
raxml_HKY = reconstruct_tree.RAxML()
start_time = time.time()
tree_rec = raxml_HKY(observations, taxa_meta, raxml_args="-T 2 --HKY85 -c 1")
if method == "SNJ":
snj = reconstruct_tree.SpectralNeighborJoining(reconstruct_tree.HKY_similarity_matrix)
start_time = time.time()
tree_rec = snj(observations, taxa_meta)
if method == "NJ":
nj = reconstruct_tree.NeighborJoining(reconstruct_tree.HKY_similarity_matrix)
start_time = time.time()
tree_rec = nj(observations, taxa_meta)
if method == "STR+NJ":
spectral_method = reconstruct_tree.STDR(reconstruct_tree.NeighborJoining, reconstruct_tree.HKY_similarity_matrix)
start_time = time.time()
tree_rec = spectral_method.deep_spectral_tree_reconstruction(observations, reconstruct_tree.HKY_similarity_matrix,
taxa_metadata = taxa_meta,
threshhold = threshold, min_split = 5, verbose = verbose)
if method == "STR+SNJ":
spectral_method = reconstruct_tree.STDR(reconstruct_tree.SpectralNeighborJoining, reconstruct_tree.HKY_similarity_matrix)
start_time = time.time()
tree_rec = spectral_method.deep_spectral_tree_reconstruction(observations, reconstruct_tree.HKY_similarity_matrix,
taxa_metadata = taxa_meta,
threshhold = threshold, min_split = 5, verbose = verbose)
if method == "STR+RaXML":
spectral_method = reconstruct_tree.STDR(reconstruct_tree.RAxML, reconstruct_tree.HKY_similarity_matrix)
start_time = time.time()
tree_rec = spectral_method.deep_spectral_tree_reconstruction(observations, reconstruct_tree.HKY_similarity_matrix,
taxa_metadata = taxa_meta,
threshhold = threshold,
raxml_args = "-T 2 --HKY85 -c 1", min_split = 5, verbose = verbose)
runtime = time.time() - start_time
RF,F1 = reconstruct_tree.compare_trees(tree_rec, tree)
print(method)
if threshold is not None: print(threshold)
print("--- %s seconds ---" % runtime)
print("RF = ",RF)
print("F1% = ",F1)
return([method, str(threshold), runtime, RF, F1])
reconstruct_tree.JC_similarity_matrix)
nj = reconstruct_tree.NeighborJoining(reconstruct_tree.JC_similarity_matrix)
treesvd = reconstruct_tree.TreeSVD()
methods = [snj, treesvd]
#results = compare_methods.experiment([reference_tree], jc, N_vec, methods=methods,\
# mutation_rates = [mutation_rate], reps_per_tree=num_reps)
df = pd.DataFrame(
columns=['method', 'runtime', 'RF', 'n', 'cherries_ref', 'cherries_res'])
for i in np.arange(num_reps):
print(i)
reference_tree = utils.unrooted_pure_kingman_tree(num_taxa)
ch_ref = cherry_count_for_tree(reference_tree)
for n in N_vec:
observations, taxa_meta = generation.simulate_sequences(
n,
tree_model=reference_tree,
seq_model=jc,
mutation_rate=mutation_rate,
alphabet="DNA")
# Tree svd
t_s = time.time()
tree_svd_rec = treesvd(observations, taxa_meta)
runtime_treesvd = time.time() - t_s
RF_svd, F1 = reconstruct_tree.compare_trees(tree_svd_rec,
reference_tree)
ch_svd = cherry_count_for_tree(tree_svd_rec)
df = df.append(
{
'method': 'treesvd',
'runtime': runtime_treesvd,
'RF': RF_svd,
threshold = 32
# jc = generation.Jukes_Cantor()
hky = generation.HKY(kappa=2)
mutation_rate = 0.05
# mutation_rate = [jc.p2t(0.95)]
np.random.seed(0)
# reference_tree = utils.unrooted_birth_death_tree(num_taxa, birth_rate=1)
# reference_tree = utils.lopsided_tree(num_taxa)
reference_tree = utils.balanced_binary(num_taxa)
# for x in reference_tree.preorder_edge_iter():
# x.length = 1
np.random.seed(0)
t0 = time.time()
observations, meta = generation.simulate_sequences(N,
tree_model=reference_tree,
seq_model=hky,
mutation_rate=mutation_rate,
rng=np.random,
alphabet='DNA')
print("gen time: ", time.time() - t0)
spectral_method = str.STR(reconstruct_tree.RAxML,
reconstruct_tree.HKY_similarity_matrix,
threshold=threshold,
merge_method="least_square",
num_gaps=1,
min_split=5,
verbose=False)
t0 = time.time()
#cProfile.run("""tree_rec = spectral_method.deep_spectral_tree_reconstruction(observations, reconstruct_tree.JC_similarity_matrix,
# taxa_metadata= meta,
# threshhold = 8, min_split = 3 ,verbose=False)""", filename="temp1.prof")
def run_method(method,
size,
run_num,
tree,
m=300,
kappa=2,
mutation_rate=0.05,
threshold=None,
verbose=False):
subtree_folder = "/gpfs/ysm/scratch60/morgan_levine/mw957/tree_merge_test/seqlen_" + str(
m) + "_" + method + "_" + str(threshold) + "_" + str(run_num) + "/"
if os.path.exists(subtree_folder):
shutil.rmtree(subtree_folder)
os.mkdir(subtree_folder)
tree.write(path=subtree_folder + "true_tree.txt", schema="newick")
subtree_filename = subtree_folder + "subtree_%s.txt"
start_time = time.time()
observations, taxa_meta = generation.simulate_sequences(
m,
tree_model=tree,
seq_model=generation.Jukes_Cantor(),
mutation_rate=mutation_rate,
alphabet="DNA")
runtime = time.time() - start_time
print("Simulation took %s seconds" % runtime)
spectral_method = reconstruct_tree.STDR(
reconstruct_tree.RAxML, reconstruct_tree.JC_similarity_matrix)
start_time = time.time()
tree_rec = spectral_method.deep_spectral_tree_reconstruction(
observations,
reconstruct_tree.JC_similarity_matrix,
taxa_metadata=taxa_meta,
threshhold=threshold,
raxml_args="-T 2 --HKY85 -c 1",
min_split=5,
verbose=verbose,
subtree_filename=subtree_filename)
runtime = time.time() - start_time
tree_rec.write(path=subtree_folder + "STDR_tree.txt", schema="newick")
distance = reconstruct_tree.JC_distance_matrix(observations, taxa_meta)
distance_pd = pd.DataFrame(distance)
taxa_list = [x.label for x in taxa_meta]
with open(subtree_folder + 'taxa.txt', 'w') as f:
for item in taxa_list:
f.write("%s\n" % item)
distance_pd.index = taxa_list
distance_path = subtree_folder + "HKY_distance.txt"
distance_pd.to_csv(distance_path, sep="\t", header=False)
with open(distance_path, 'r') as original:
data = original.read()
with open(distance_path, 'w') as modified:
modified.write(str(size) + "\n" + data)
# accuracy of the STDR method
RF, F1 = reconstruct_tree.compare_trees(tree_rec, tree)
print(method)
if threshold is not None: print(threshold)
print("--- %s seconds ---" % runtime)
print("RF = ", RF)
print("F1% = ", F1)
return ([method, str(threshold), runtime, RF, F1])
jc = generation.Jukes_Cantor()
mutation_rate = [jc.p2t(0.95)]
num_itr = 2 #0
# reference_tree = utils.unrooted_birth_death_tree(num_taxa, birth_rate=1)
# for x in reference_tree.preorder_edge_iter():
# x.length = 1
merging_method_list = ['least_square', 'angle']
RF = {'least_square': [], 'angle': []}
F1 = {'least_square': [], 'angle': []}
for merge_method in merging_method_list:
for i in range(num_itr):
#reference_tree = utils.balanced_binary(num_taxa)
reference_tree = utils.lopsided_tree(num_taxa)
observations, taxa_meta = generation.simulate_sequences(
N,
tree_model=reference_tree,
seq_model=jc,
mutation_rate=mutation_rate)
spectral_method = reconstruct_tree.SpectralTreeReconstruction(
reconstruct_tree.NeighborJoining,
reconstruct_tree.JC_similarity_matrix)
tree_rec = spectral_method.deep_spectral_tree_reconstruction(
observations,
reconstruct_tree.JC_similarity_matrix,
taxa_metadata=taxa_meta,
threshhold=16,
merge_method=merge_method)
RF_i, F1_i = reconstruct_tree.compare_trees(tree_rec, reference_tree)
RF[merge_method].append(RF_i)
F1[merge_method].append(F1_i)