def test_procrustes_monte_carlo(self):
""" sanity test of procrustes_monte_carlo wrapper function
THIS TEST MAY RANDOMLY FAIL BECAUSE IT IS BASED ON TESTING
RANDOM PERMUTATIONS, BUT THAT SHOULD BE RARE.
"""
def shuffle_f(coords):
""" A fake shuffle function -- used to avoid random test failures
returns a re-ordered coords2
"""
return array(
[
[-0.16713783, 0.22321481, 0.33766418, 0.22785083, -0.23830446, -0.18754852],
[-0.14864343, 0.07290181, -0.06250315, 0.03201972, -0.0966749, -0.10337987],
[0.35725269, -0.00761567, 0.09044279, -0.21006839, -0.01355589, -0.04590791],
[0.26535811, 0.09772598, 0.04339214, -0.21014987, 0.14089095, -0.10261849],
]
)
actual = procrustes_monte_carlo(StringIO(pcoa1_f), StringIO(pcoa2_f), trials=100, shuffle_f=shuffle_f)
# just some sanity checks as the individual componenets are
# already tested -- these are based on looking at the output of the
# run, and testing to ensure that it hasn't changed
expected_actual_m2 = 0.0211
expected_len_trial_m2 = 100
expected_count_better = 0
expected_p_value = 0.0
self.assertAlmostEqual(actual[0], expected_actual_m2, 3)
self.assertEqual(len(actual[1]), expected_len_trial_m2)
self.assertEqual(actual[2], expected_count_better)
self.assertEqual(actual[3], expected_p_value)
def test_procrustes_monte_carlo(self):
""" sanity test of procrustes_monte_carlo wrapper function
THIS TEST MAY RANDOMLY FAIL BECAUSE IT IS BASED ON TESTING
RANDOM PERMUTATIONS, BUT THAT SHOULD BE RARE.
"""
def shuffle_f(coords):
""" A fake shuffle function -- used to avoid random test failures
returns a re-ordered coords2
"""
return array(\
[[-0.16713783, 0.22321481, 0.33766418, 0.22785083, -0.23830446, -0.18754852],
[-0.14864343, 0.07290181, -0.06250315, 0.03201972, -0.0966749 , -0.10337987],
[ 0.35725269, -0.00761567, 0.09044279, -0.21006839, -0.01355589, -0.04590791],
[ 0.26535811, 0.09772598, 0.04339214, -0.21014987, 0.14089095, -0.10261849]])
actual = procrustes_monte_carlo(self.pcoa1_f,
self.pcoa2_f,
trials=100,
shuffle_f=shuffle_f)
# just some sanity checks as the individual componenets are
# already tested -- these are based on looking at the output of the
# run, and testing to ensure that it hasn't changed
expected_actual_m2 = 0.0211
expected_len_trial_m2 = 100
expected_count_better = 0
expected_p_value = 0.0
self.assertAlmostEqual(actual[0], expected_actual_m2, 3)
self.assertEqual(len(actual[1]), expected_len_trial_m2)
self.assertEqual(actual[2], expected_count_better)
self.assertEqual(actual[3], expected_p_value)
def compute_procrustes(result_tables, expected_pc_lookup, taxonomy_level=6, num_dimensions=3, random_trials=999):
""" Compute Procrustes M2 and p-values for a set of results
result_tables: 2d list of tables to be compared to expected tables,
where the data in the inner list is:
[dataset_id, reference_database_id, method_id,
parameter_combination_id, table_fp]
expected_pc_lookup: 2d dict of dataset_id, reference_db_id to principal
coordinate matrices, for the expected result coordinate matrices
taxonomy_level: level to compute results
"""
### Start code copied ALMOST* directly from compute_prfs - some re-factoring for re-use is
### in order here. *ALMOST refers to changes to parser and variable names since expected
### is a pc matrix here.
for dataset_id, reference_id, method_id, params, actual_table_fp in result_tables:
## parse the expected table (unless taxonomy_level is specified, this should be
## collapsed on level 6 taxonomy)
try:
expected_pc_fp = expected_pc_lookup[dataset_id][reference_id]
except KeyError:
raise KeyError, "Can't find expected table for (%s, %s)." % (dataset_id, reference_id)
## parse the actual table and collapse it at the specified taxonomic level
try:
actual_table = parse_biom_table(open(actual_table_fp, "U"))
except ValueError:
raise ValueError, "Couldn't parse BIOM table: %s" % actual_table_fp
collapse_by_taxonomy = get_taxonomy_collapser(taxonomy_level)
actual_table = actual_table.collapseObservationsByMetadata(collapse_by_taxonomy)
### End code copied directly from compute_prfs.
# Next block of code, how do I hate thee? Let me count the ways...
# (1) dist_bray_curtis doesn't take a BIOM Table object
# (2) pcoa takes a qiime-formatted distance matrix as a list of lines
# (3) pcoa return a qiime-formatted pc matrix
# (4) procrustes_monte_carlo needs to pass through the pc "file" multiple
# times, so we actually *need* those the pcs that get passed in to be
# lists of lines
dm = dist_bray_curtis(asarray([v for v in actual_table.iterSampleData()]))
formatted_dm = format_distance_matrix(actual_table.SampleIds, dm)
actual_pc = pcoa(formatted_dm.split("\n")).split("\n")
expected_pc = list(open(expected_pc_fp, "U"))
## run Procrustes analysis with monte carlo simulation
actual_m_squared, trial_m_squareds, count_better, mc_p_value = procrustes_monte_carlo(
expected_pc,
actual_pc,
trials=random_trials,
max_dimensions=num_dimensions,
sample_id_map=None,
trial_output_dir=None,
)
yield (dataset_id, reference_id, method_id, params, actual_m_squared, mc_p_value)
def test_procrustes_monte_carlo(self):
""" sanity test of procrustes_monte_carlo wrapper function"""
actual = procrustes_monte_carlo(self.pcoa1_f,self.pcoa2_f,trials=10)
# just some sanity checks as the individual componenets are
# already tested -- these are based on looking at the output of the
# run, and testing to ensure that it hasn't changed
expected_actual_m2 = 0.0211
expected_len_trial_m2 = 10
expected_count_better = 0
expected_p_value = 0.0
self.assertAlmostEqual(actual[0],expected_actual_m2,3)
self.assertEqual(len(actual[1]),expected_len_trial_m2)
self.assertEqual(actual[2],expected_count_better)
self.assertEqual(actual[3],expected_p_value)
def test_procrustes_monte_carlo(self):
""" sanity test of procrustes_monte_carlo wrapper function"""
actual = procrustes_monte_carlo(self.pcoa1_f, self.pcoa2_f, trials=10)
# just some sanity checks as the individual componenets are
# already tested -- these are based on looking at the output of the
# run, and testing to ensure that it hasn't changed
expected_actual_m2 = 0.0211
expected_len_trial_m2 = 10
expected_count_better = 0
expected_p_value = 0.0
self.assertAlmostEqual(actual[0], expected_actual_m2, 3)
self.assertEqual(len(actual[1]), expected_len_trial_m2)
self.assertEqual(actual[2], expected_count_better)
self.assertEqual(actual[3], expected_p_value)
def main():
option_parser, opts, args = parse_command_line_parameters(**script_info)
input_fps = opts.input_fps
sample_id_map_fps = opts.sample_id_map_fps
num_dimensions = opts.num_dimensions
max_dims_str = str(num_dimensions or 'alldim')
output_dir = opts.output_dir
random_trials = opts.random_trials
if random_trials is not None and random_trials < 10:
option_parser.error(
'Must perform >= 10 trails for Monte Carlo analysis.')
if sample_id_map_fps and \
(len(sample_id_map_fps) + 1) != len(opts.input_fps):
option_parser.error('If providing sample id maps, there must be '
'exactly one fewer sample id maps than input '
'coordinate matrices.')
if not exists(output_dir):
makedirs(output_dir)
reference_input_fp = input_fps[0]
reference_input_fp_dir, input_fn1 = split(reference_input_fp)
reference_input_fp_basename, reference_input_fp_ext = splitext(input_fn1)
output_summary_fp = join(output_dir, 'procrustes_results.txt')
summary_file_lines = [
'#FP1\tFP2\tNum included dimensions\tMonte Carlo '
'p-value\tCount better\tM^2',
'#Warning: p-values in this file are NOT currently '
'adjusted for multiple comparisons.'
]
for i, query_input_fp in enumerate(input_fps[1:]):
query_input_fp_dir, query_input_fn = split(query_input_fp)
query_input_fp_basename, query_input_fp_ext = splitext(query_input_fn)
output_matrix1_fp = join(
output_dir,
'%s_transformed_reference.txt' % reference_input_fp_basename)
output_matrix2_fp = join(
output_dir,
'%s_transformed_q%d.txt' % (query_input_fp_basename, i + 1))
if sample_id_map_fps:
with open(sample_id_map_fps[i], "U") as f:
sample_id_map = dict([
(k, v[0]) for k, v in fields_to_dict(f).iteritems()
])
else:
sample_id_map = None
with open(reference_input_fp, 'U') as ref_in_f:
with open(query_input_fp, 'U') as query_in_f:
transf_coords1, transf_coords2, m_squared, rand_coords2 =\
get_procrustes_results(ref_in_f, query_in_f,
sample_id_map=sample_id_map,
randomize=False,
max_dimensions=num_dimensions)
transf_coords1.write(output_matrix1_fp)
transf_coords2.write(output_matrix2_fp)
if random_trials:
if opts.store_trial_details:
trial_output_dir = join(output_dir, 'trial_details_%d' % i + 2)
else:
trial_output_dir = None
coords_f1 = open(reference_input_fp, 'U')
coords_f2 = open(query_input_fp, 'U')
actual_m_squared, trial_m_squareds, count_better, mc_p_value =\
procrustes_monte_carlo(coords_f1,
coords_f2,
trials=random_trials,
max_dimensions=num_dimensions,
sample_id_map=sample_id_map,
trial_output_dir=trial_output_dir)
# truncate the p-value to the correct number of significant
# digits
mc_p_value_str = format_p_value_for_num_iters(
mc_p_value, random_trials)
summary_file_lines.append(
'%s\t%s\t%s\t%s\t%d\t%1.3f' %
(reference_input_fp, query_input_fp, max_dims_str,
mc_p_value_str, count_better, actual_m_squared))
else:
summary_file_lines.append(
'%s\t%s\t%s\tNA\tNA\t%1.3f' %
(reference_input_fp, query_input_fp, max_dims_str, m_squared))
# Write output summary
with open(output_summary_fp, 'w') as f:
f.write('\n'.join(summary_file_lines))
f.write('\n')
def main():
option_parser, opts, args = parse_command_line_parameters(**script_info)
input_fps = opts.input_fps
sample_id_map_fps = opts.sample_id_map_fps
num_dimensions = opts.num_dimensions
max_dims_str = str(num_dimensions or 'alldim')
output_dir = opts.output_dir
random_trials = opts.random_trials
if random_trials != None and random_trials < 10:
option_parser.error('Must perform >= 10 trails for Monte Carlo analysis.')
if sample_id_map_fps and \
(len(sample_id_map_fps) + 1) != len(opts.input_fps):
option_parser.error('If providing sample id maps, there must be exactly'
' one fewer sample id maps than input coordinate'
' matrices.')
if not exists(output_dir):
makedirs(output_dir)
reference_input_fp = input_fps[0]
reference_input_fp_dir, input_fn1 = split(reference_input_fp)
reference_input_fp_basename, reference_input_fp_ext = splitext(input_fn1)
output_summary_fp = join(output_dir,'procrustes_results.txt')
summary_file_lines = \
['#FP1\tFP2\tNum included dimensions\tMonte Carlo p-value\tCount better\tM^2',
'#Warning: p-values in this file are NOT currently adjusted for multiple comparisons.']
for i,query_input_fp in enumerate(input_fps[1:]):
query_input_fp_dir, query_input_fn = split(query_input_fp)
query_input_fp_basename, query_input_fp_ext = splitext(query_input_fn)
output_matrix1_fp = join(output_dir,
'%s_transformed_reference.txt' % reference_input_fp_basename)
output_matrix2_fp = join(output_dir,\
'%s_transformed_q%d.txt' % (query_input_fp_basename, i+1))
if sample_id_map_fps:
sample_id_map = dict([(k,v[0]) \
for k,v in fields_to_dict(open(sample_id_map_fps[i], "U")).items()])
else:
sample_id_map = None
transformed_coords1, transformed_coords2, m_squared, randomized_coords2 =\
get_procrustes_results(open(reference_input_fp,'U'),\
open(query_input_fp,'U'),\
sample_id_map=sample_id_map,\
randomize=False,
max_dimensions=num_dimensions)
output_matrix1_f = open(output_matrix1_fp,'w')
output_matrix1_f.write(transformed_coords1)
output_matrix1_f.close()
output_matrix2_f = open(output_matrix2_fp,'w')
output_matrix2_f.write(transformed_coords2)
output_matrix2_f.close()
if random_trials:
if opts.store_trial_details:
trial_output_dir = join(output_dir,'trial_details_%d' % i+2)
else:
trial_output_dir = None
coords_f1 = list(open(reference_input_fp,'U'))
coords_f2 = list(open(query_input_fp,'U'))
actual_m_squared, trial_m_squareds, count_better, mc_p_value =\
procrustes_monte_carlo(coords_f1,
coords_f2,
trials=random_trials,
max_dimensions=num_dimensions,
sample_id_map=sample_id_map,
trial_output_dir=trial_output_dir)
# truncate the p-value to the correct number of significant
# digits
mc_p_value_str = format_p_value_for_num_iters(mc_p_value, random_trials)
summary_file_lines.append('%s\t%s\t%s\t%s\t%d\t%1.3f' %\
(reference_input_fp, query_input_fp, max_dims_str, mc_p_value_str,\
count_better, actual_m_squared))
else:
summary_file_lines.append('%s\t%s\t%s\tNA\tNA\t%1.3f' %\
(reference_input_fp, query_input_fp, max_dims_str, m_squared))
# Write output summary
f = open(output_summary_fp,'w')
f.write('\n'.join(summary_file_lines))
f.write('\n')
f.close()
def main():
option_parser, opts, args = parse_command_line_parameters(**script_info)
random_trials = opts.random_trials
if random_trials != None and random_trials < 10:
option_parser.error('Must perform >= 10 trails for Monte Carlo analysis.')
output_dir = opts.output_dir
sample_id_map_fp = opts.sample_id_map_fp
num_dimensions = opts.num_dimensions
if not exists(output_dir):
makedirs(output_dir)
if opts.store_trial_details:
trial_output_dir = '%s/trial_details/' % output_dir
else:
trial_output_dir = None
input_fp1 = opts.input_fps[0]
input_fp2 = opts.input_fps[1]
input_fp1_dir, input_fn1 = split(input_fp1)
input_fp1_basename, input_fp1_ext = splitext(input_fn1)
input_fp2_dir, input_fn2 = split(input_fp2)
input_fp2_basename, input_fp2_ext = splitext(input_fn2)
output_summary_fp = '%s/%s_%s_procrustes_results.txt' %\
(output_dir,input_fp1_basename,input_fp2_basename)
output_matrix1_fp = '%s/pc1_transformed.txt' % output_dir
output_matrix2_fp = '%s/pc2_transformed.txt' % output_dir
if sample_id_map_fp:
sample_id_map = dict([(k,v[0]) \
for k,v in fields_to_dict(open(sample_id_map_fp, "U")).items()])
else:
sample_id_map = None
transformed_coords1, transformed_coords2, m_squared, randomized_coords2 =\
get_procrustes_results(open(input_fp1,'U'),\
open(input_fp2,'U'),\
sample_id_map=sample_id_map,\
randomize=False,
max_dimensions=num_dimensions)
output_matrix1_f = open(output_matrix1_fp,'w')
output_matrix1_f.write(transformed_coords1)
output_matrix1_f.close()
output_matrix2_f = open(output_matrix2_fp,'w')
output_matrix2_f.write(transformed_coords2)
output_matrix2_f.close()
if random_trials:
summary_file_lines = ['FP1 FP2 Included_dimensions MC_p_value Count_better M^2']
coords_f1 = list(open(input_fp1,'U'))
coords_f2 = list(open(input_fp2,'U'))
actual_m_squared, trial_m_squareds, count_better, mc_p_value =\
procrustes_monte_carlo(coords_f1,\
coords_f2,\
trials=random_trials,\
max_dimensions=num_dimensions,
sample_id_map=sample_id_map,
trial_output_dir=trial_output_dir)
# truncate the p-value to the correct number of significant
# digits
mc_p_value_str = format_p_value_for_num_iters(mc_p_value, random_trials)
max_dims_str = str(num_dimensions or 'alldim')
summary_file_lines.append('%s %s %s %s %d %1.3f' %\
(input_fp1, input_fp2, str(max_dims_str), mc_p_value_str,\
count_better, actual_m_squared))
f = open(output_summary_fp,'w')
f.write('\n'.join(summary_file_lines))
f.write('\n')
f.close()
