def subsample_neighbors(focal_seqs: DNAFASTAFormat,
context_seqs: DNAFASTAFormat,
percent_id: float,
samples_per_cluster: int,
locale: CategoricalMetadataColumn = None,
max_accepts: int = 10,
n_threads: int = 1,
seed: int = None) -> IDSelection:
if max_accepts < samples_per_cluster:
raise ValueError('max_accepts (%d) must be greater than or equal to '
'samples_per_cluster (%d), since it is determines '
'the largest number of samples that could be '
'obtained per cluster.' %
(max_accepts, samples_per_cluster))
context_ids = ids_from_fasta(str(context_seqs))
inclusion = pd.Series(False, index=context_ids, name='inclusion')
if locale is not None:
locale = locale.filter_ids(inclusion.index).to_series()
metadata = pd.DataFrame(locale)
else:
metadata = pd.DataFrame(index=pd.Index(inclusion.index))
metadata.index.name = 'id'
with tempfile.NamedTemporaryFile() as vsearch_out_f:
command = [
'vsearch', '--threads',
str(n_threads), '--usearch_global',
str(focal_seqs), '--id',
str(percent_id), '--db',
str(context_seqs), '--userout', vsearch_out_f.name, '--qmask',
'none', '--maxaccepts',
str(max_accepts), '--uc_allhits', '--userfields',
'query+target+mism'
]
run_command(command)
vsearch_out = pd.read_csv(
open(vsearch_out_f.name),
sep='\t',
na_values='*',
names=['focal_id', 'context_id', 'n_mismatches'])
clusters = _clusters_from_vsearch_out(vsearch_out, locale)
context_seqs_to_keep = \
_sample_clusters(clusters, samples_per_cluster, seed=seed)
inclusion[context_seqs_to_keep] = True
return IDSelection(inclusion, qiime2.Metadata(metadata),
"subsample_neighbors")
def subsample_diversity(context_seqs: DNAFASTAFormat,
percent_id: float,
max_accepts: int = 10,
n_threads: int = 1) -> IDSelection:
context_ids = ids_from_fasta(str(context_seqs))
inclusion = pd.Series(False, index=context_ids, name='inclusion')
metadata = pd.DataFrame(index=pd.Index(inclusion.index))
metadata.index.name = 'id'
with tempfile.NamedTemporaryFile() as uc_out_f:
command = [
'vsearch',
'--threads',
str(n_threads),
'--cluster_fast',
str(context_seqs),
'--id',
str(percent_id),
'--uc',
uc_out_f.name,
'--qmask',
'none',
'--maxaccepts',
str(max_accepts),
]
run_command(command)
uc = pd.read_csv(uc_out_f.name,
sep='\t',
na_values='*',
names=[
'type', 'cluster_id', 'length', 'perc_id',
'strand', 'BLANK1', 'BLANK2', 'cigar', 'query',
'target'
])
# the S lines define the cluster centroids
context_seqs_to_keep = uc[uc['type'] == 'S'].index
inclusion[context_seqs_to_keep] = True
return IDSelection(inclusion, qiime2.Metadata(metadata),
"subsample_diversity")
def sample_longitudinal(dates: qiime2.CategoricalMetadataColumn,
context_seqs: DNAFASTAFormat = None,
start_date: str = None,
samples_per_interval: int = 7,
days_per_interval: int = 7,
seed: int = None) -> IDSelection:
window_size = '%dD' % days_per_interval
if context_seqs is not None:
# filter dates to only include the ids that sequence data is
# available for
ids_to_include = ids_from_fasta(str(context_seqs))
dates = dates.filter_ids(ids_to_include)
dt_series = pd.to_datetime(dates.to_series(), errors='coerce')
df = pd.DataFrame({'ids': dates.to_series().index}, index=dt_series)
if start_date is not None:
filter_before = pd.Timestamp(start_date)
df = df.iloc[np.where(dt_series >= filter_before)]
if filter_before not in df.index:
# this will be stripped in _sample_group::_sampler
# the purpose is to force Pandas to begin the window at this
# time instead of the first observation (by making NaN the first
# observation)
df.loc[filter_before] = float('nan')
grouped = df.groupby(pd.Grouper(freq=window_size,
convention='start',
closed='left'),
group_keys=False)
filtered_df = grouped.apply(_sample_group(samples_per_interval, seed))
df = df.dropna(axis=0)
selection = pd.Series(False, index=dates.to_series().index)
selection[filtered_df['ids']] = True
md = qiime2.Metadata(dates.to_dataframe())
return IDSelection(selection, md, 'sample_longitudinal')