def kmer2distance(testset_dir, num_testset, size_testset, size_phy, k_mer=6):
for i in range(num_testset):
print('testset[{}] calculate the distance matrix ...'.format(i + 1))
for j in range(size_testset):
msa_file = os.path.join(testset_dir, str(i + 1),
'%d_rn.fasta' % (j + 1))
out_file = os.path.join(testset_dir, str(i + 1),
'%d_dist.meg' % (j + 1))
with open(msa_file, 'rt') as f_read:
seq_list = [seq for _, seq in SimpleFastaParser(f_read)]
vecs = np.array([get_kmer(seq, k_mer=k_mer) for seq in seq_list])
dist = euclidean_distances(vecs, vecs)
with open(out_file, 'wt') as f_write:
# mega distance format
f_write.write('#mega\n')
f_write.write(
'!TITLE cosine distance matrix of k-mer vectors;\n')
f_write.write(
'!Format DataType=distance NTaxa={};\n'.format(size_phy))
f_write.write('\n')
f_write.write('\n'.join('#{}'.format(k + 1)
for k in range(size_phy)))
f_write.write('\n\n\n')
# distance matrix
for r in range(1, size_phy):
f_write.write(' '.join(str(a) for a in dist[r, :r]))
f_write.write('\n')
def _get_kmer_vecs(self, seqs_batch, k_mer=6):
'''Get k-mer representation of sequences.
param:
seqs_batch: list of sequence list
'''
vecs_batch = [[get_kmer(seq, k_mer) for seq in seqs]
for seqs in seqs_batch]
return vecs_batch
def infer_kmer_vector(seq_file, pickle_file, k_mer=5):
print('Inferring k-mer vector with k={}'.format(k_mer))
with open(seq_file, 'rt') as f_read:
vectors = [
get_kmer(seq, k_mer=k_mer)
for _, seq in tqdm(SimpleFastaParser(f_read))
]
vectors = torch.tensor(vectors, dtype=torch.float)
torch.save(vectors, pickle_file)
def compare_variance(method_types,
testset_dir,
num_testset,
size_testset,
size_phy,
out_filename=None,
k_mer=6,
silent=False):
if out_filename is None:
out_filename = 'variance.csv'
total_result_file = os.path.join(testset_dir, out_filename)
total_results = {method_type: [] for method_type in method_types}
for i in range(num_testset):
print('testset[{}] Calculating varience ...'.format(i + 1))
results = {method_type: [] for method_type in method_types}
for j in range(size_testset):
# distance matrix
msa_file = os.path.join(testset_dir, str(i + 1),
'%d_rn.fasta' % (j + 1))
with open(msa_file, 'rt') as f_read:
seq_list = [seq for _, seq in SimpleFastaParser(f_read)]
vecs = np.array([get_kmer(seq, k_mer=k_mer) for seq in seq_list])
dist = euclidean_distances(vecs, vecs)
for method_type in method_types:
tree_file = os.path.join(testset_dir, str(i + 1),
'%d_%s.nwk' % (j + 1, method_type))
tree = TreeIO.read(tree_file, 'newick')
# print('i: {},j: {}, method_type: {}'.format(i, j, method_type))
var = cal_tree_varience(tree, dist, size_phy)
results[method_type].append(var)
avg = {
method_type: np.mean(results[method_type])
for method_type in method_types
}
for method_type in method_types:
total_results[method_type].append(avg[method_type])
result_file = os.path.join(testset_dir, str(i + 1), out_filename)
with open(result_file, 'wt') as f_write:
for method_type in method_types:
ss = ['{:.2f}'.format(num) for num in results[method_type]]
f_write.write('{},{}\n'.format(method_type, ','.join(ss)))
with open(total_result_file, 'wt') as f_write:
for method_type in method_types:
ss = ['{:.2f}'.format(num) for num in total_results[method_type]]
f_write.write('{},{}\n'.format(method_type, ','.join(ss)))
def __init__(self, seq_file=None, vec_path=None, k_mer=5, seq2vec=None):
# assert isinstance(seq2vec, Doc2Vec), 'seq2vec must be instance of Doc2Vec'
# self._seq2vec = seq2vec
# self.infer_vec = self._seq2vec.infer_vector
if vec_path is not None:
self.vectors = torch.load(vec_path)
elif seq_file is not None:
with open(seq_file, 'rt') as f_read:
self.vectors = torch.tensor([get_kmer(seq, k_mer=k_mer)
for _, seq in SimpleFastaParser(f_read)],
dtype=torch.float)
else:
raise RuntimeError('vec_path or seq_file is required!')
def infer_tree_tsp(testset_dir, num_testset, size_testset, k_mer=6):
for i in range(num_testset):
print('testset[{}] Inferring tree by tsp ...'.format(i + 1))
for j in range(size_testset):
msa_file = os.path.join(testset_dir, str(i + 1),
'%d_rn.fasta' % (j + 1))
out_file = os.path.join(testset_dir, str(i + 1),
'%d_tsp.nwk' % (j + 1))
with open(msa_file, 'rt') as f_read:
seq_list = [seq for _, seq in SimpleFastaParser(f_read)]
vecs = [get_kmer(seq, k_mer=k_mer) for seq in seq_list]
co = gurobi_tsp(vecs)
tree = cluster_co2tree(vecs, co)
# tree = co2tree(vecs, co)
TreeIO.write([tree], out_file, 'newick')
def infer_tree_rl(model_path,
node_dim,
embedding_dim,
n_encode_layers,
testset_dir,
num_testset,
size_testset,
k_mer=6,
out_filename=''):
model = AttentionModel(node_dim=node_dim,
embedding_dim=embedding_dim,
n_encode_layers=n_encode_layers)
print(' [*] Loading data from {}'.format(model_path))
load_data = torch_load_cpu(model_path)
model.load_state_dict(load_data['model'])
# del load_data
model.set_decode_type('greedy')
for i in range(num_testset):
print('testset[{}] Inferring tree by rl ...'.format(i + 1))
for j in range(size_testset):
msa_file = os.path.join(testset_dir, str(i + 1),
'%d_rn.fasta' % (j + 1))
out_file = os.path.join(testset_dir, str(i + 1),
'%d_rl%s.nwk' % (j + 1, out_filename))
with open(msa_file, 'rt') as f_read:
seq_list = [seq for _, seq in SimpleFastaParser(f_read)]
vecs = [get_kmer(seq, k_mer=k_mer) for seq in seq_list]
vecs_tensor = torch.tensor(vecs, dtype=torch.float).unsqueeze(0)
_, _, circular_orders = model(vecs_tensor, return_pi=True)
co = circular_orders.tolist()[0]
tree = co2tree(vecs, co)
# tree = cluster_co2tree(vecs, co)
TreeIO.write([tree], out_file, 'newick')