def __init__(self, fastas, labels, max_seqs=45):
"""
Creates dataset object from given alignments. Drops ones with len>max_lens.
You can create an empty dataset to load your data later. It is more memory-friendly approach.
:param fastas: alignment files
:param labels: labels of each fasta
:param max_seqs: maximum number of sequences in one file
"""
self.is_sparse = False
self.edge_indices = None
self.edge_weights = None
self.f = None # holder for hdf5 file
self.data = np.zeros((len(fastas), max_seqs, 3))
self.labels = np.array(labels, dtype=np.long)
self.dist = np.zeros((len(fastas), max_seqs, max_seqs))
calc = TreeConstruction.DistanceCalculator(
'blosum62') # calculator for distance matrices
for file_index, file in enumerate(fastas):
aln = AlignIO.read(file, "fasta")
for seq_index, seq in enumerate(aln):
for col_index, col in enumerate(seq):
self.data[file_index,
seq_index] += aesnn3[res_to_index[col]]
self.dist[file_index, :len(aln), :len(aln)] = np.array(
calc.get_distance(aln))
def labeler(files, etalon_tree, tree_path=".", rebuild=False):
"""
Constructs labels for given files. (Best phylogeny reconstruction method)
:param files: an iterable with file paths to alignments
:param etalon_tree: the path to etalon tree
:param tree_path: a directory, where built trees will be stored
:param rebuild: set it True, if you need to rebuild trees or build them from scratch
:return: tensor with labels
"""
tree_path = osp.abspath(tree_path) # raxml needs absolute paths
if rebuild:
calculator = TreeConstruction.DistanceCalculator('blosum62')
dist_constructor = TreeConstruction.DistanceTreeConstructor()
# construct all trees with UPGMA, NJ and raxml
for i, file in enumerate(files):
aln = AlignIO.read(file, 'fasta')
tree = dist_constructor.upgma(calculator.get_distance(aln))
name = file.split("/")[-1].split(".")[0]
Phylo.write(tree, osp.join(tree_path, 'upgma_{}.tre'.format(name)),
'newick')
tree = dist_constructor.nj(calculator.get_distance(aln))
Phylo.write(tree, osp.join(tree_path, 'nj_{}.tre'.format(name)),
'newick')
raxml = RaxmlCommandline(sequences=osp.abspath(file),
model='PROTCATWAG',
name='{}.tre'.format(name),
threads=3,
working_dir=tree_path)
_, stderr = raxml()
print(stderr)
print('{} finished'.format(name))
# get best tree
tns = dendropy.TaxonNamespace()
act_tree = dendropy.Tree.get_from_path(osp.join(tree_path, etalon_tree),
"newick",
taxon_namespace=tns)
act_tree.encode_bipartitions()
distances = np.zeros(shape=(len(files), 3))
for i, file in enumerate(files):
name = file.split("/")[-1].split(".")[0]
nj_tree = dendropy.Tree.get_from_path(osp.join(
tree_path, "nj_{}.tre".format(name)),
"newick",
taxon_namespace=tns)
up_tree = dendropy.Tree.get_from_path(osp.join(
tree_path, "upgma_{}.tre".format(name)),
"newick",
taxon_namespace=tns)
ml_tree = dendropy.Tree.get_from_path(osp.join(
tree_path, "RAxML_bestTree.{}.tre".format(name)),
"newick",
taxon_namespace=tns)
distances[i, 0] = dendropy.calculate.treecompare.symmetric_difference(
nj_tree, act_tree)
distances[i, 1] = dendropy.calculate.treecompare.symmetric_difference(
up_tree, act_tree)
distances[i, 2] = dendropy.calculate.treecompare.symmetric_difference(
ml_tree, act_tree)
return distances.argmin(1)
def neighbor_joining_tree(aln, prot_model='blosum62'):
""" Estimate a tree from an alignment using neighbor-joining
"""
calculator = TreeConstruction.DistanceCalculator(prot_model)
constructor = TreeConstruction.DistanceTreeConstructor(calculator, 'nj')
tree = constructor.build_tree(aln)
return(tree)
