def testRNASeq(self):
from topslam.simulation.simulate_trajectory import rnaseq_simulation
p_dims = 30
Xsim, simulate_new, t, c, labels, seed = rnaseq_simulation(p_dims, 4, 2, 1234, split_prob=.5)
np.random.seed(42)
Y, d = simulate_new()
from topslam.filtering import filter_RNASeq
import pandas as pd
E = filter_RNASeq(pd.DataFrame(np.exp(Y)))
np.testing.assert_array_less(-E.values, 0)
E = filter_RNASeq(pd.DataFrame(np.exp(Y)), transform_log1p=False)
np.testing.assert_allclose(-Y[d].mean(), 0)
np.testing.assert_array_less((E.values==0).sum(), (Y==0).sum())
def testRNASeq(self):
from topslam.simulation.simulate_trajectory import rnaseq_simulation
p_dims = 30
Xsim, simulate_new, t, c, labels, seed = rnaseq_simulation(
p_dims, 4, 2, 1234, split_prob=.5)
np.random.seed(42)
Y, d = simulate_new()
from topslam.filtering import filter_RNASeq
import pandas as pd
E = filter_RNASeq(pd.DataFrame(np.exp(Y)))
np.testing.assert_array_less(-E.values, 0)
E = filter_RNASeq(pd.DataFrame(np.exp(Y)), transform_log1p=False)
np.testing.assert_allclose(-Y[d].mean(), 0)
np.testing.assert_array_less((E.values == 0).sum(), (Y == 0).sum())
def example_deng(optimize=True, plot=True):
import pandas as pd, os
import GPy, numpy as np
from topslam.filtering import filter_RNASeq
# Reproduceability, BGPLVM has local optima
np.random.seed(42)
# This is the process of how we loaded the data:
ulabels = ['Zygote',
'2-cell embryo',
'Early 2-cell blastomere', 'Mid 2-cell blastomere', 'Late 2-cell blastomere',
'4-cell blastomere', '8-cell blastomere', '16-cell blastomere',
'Early blastocyst cell', 'Mid blastocyst cell', 'Late blastocyst cell',
'fibroblast',
'adult liver',
]
folder_path = os.path.expanduser('~/tmp/Deng')
csv_file = os.path.join(folder_path, 'filtered_expression_values.csv')
if os.path.exists(csv_file):
print('Loading previous filtered data: {}'.format(csv_file))
Y_bgplvm = pd.read_csv(csv_file, index_col=[0,1,2], header=0)
else:
print('Loading data:')
data = GPy.util.datasets.singlecell_rna_seq_deng()
if not os.path.exists(folder_path):
os.mkdir(folder_path)
Ydata = data['Y'].copy()
Ydata.columns = Ydata.columns.to_series().apply(str.upper)
Ydata = Ydata.reset_index().set_index('index', append=True)
Ydata['labels'] = data['labels'].values
Ydata = Ydata.set_index('labels', append=True)
Ydata = Ydata.reorder_levels([0,2,1])
Ydata = Ydata.reset_index([0,2]).loc[ulabels].set_index(['level_0', 'index'], append=True)
Y = Ydata.copy()
Y.columns = [c.split('.')[0] for c in Y.columns]
Y_bgplvm = filter_RNASeq(Y)
print('\nSaving data to tmp file: {}'.format(csv_file))
Y_bgplvm.to_csv(csv_file)
labels = Y_bgplvm.index.get_level_values(0).values
Ymean = Y_bgplvm.values.mean()
Ystd = Y_bgplvm.values.std()
Y_m = Y_bgplvm.values
Y_m -= Ymean
Y_m /= Ystd
# get the labels right for split experiments
# get the labels right for 8 and split
new_8_labels = []
for _l in Y_bgplvm.loc['8-cell blastomere'].index.get_level_values(1):
_l = _l.split('-')[0]
if not('split' in _l):
new_8_labels.append('8')
elif not('pooled' in _l):
new_8_labels.append('8 split')
else:
new_8_labels.append('8 split')
labels[labels=='8-cell blastomere'] = new_8_labels
# get the labels right for 16 and split
new_16_labels = []
for _l in Y_bgplvm.loc['16-cell blastomere'].index.get_level_values(1):
_l = _l.split('-')[0]
if not('split' in _l):
new_16_labels.append('16')
elif not('pooled' in _l):
new_16_labels.append('16 split')
else:
new_16_labels.append('16 split')
labels[labels=='16-cell blastomere'] = new_16_labels
ulabels = []
for lab in labels:
if lab not in ulabels:
ulabels.append(lab)
short_labels = labels.copy()
_ulabels_convert = np.array([
'Z',# Z',
'E',# Em',
'2',# Bm E',
'2',# Bm M',
'2',# Bm L',
'4',
'8',
'8 s',
'16',
'16 s',
'Bz',# E',
'Bz',# M',
'Bz',# L'
'F',
'L'
])
short_ulabels = []
for lab, nlab in zip(ulabels, _ulabels_convert):
short_labels[short_labels==lab] = nlab
if nlab not in short_ulabels:
short_ulabels.append(nlab)
from topslam.optimization import run_methods, methods, create_model, optimize_model
X_init, dims = run_methods(Y_m, methods)
m = create_model(Y_m, X_init, num_inducing=25)
m.Ymean = Ymean
m.Ystd = Ystd
m.data_labels = short_labels
m.data_ulabels = short_ulabels
m.data = Y_bgplvm
m.X_init = X_init
m.dims = dims
if optimize:
optimize_model(m)
if plot:
mc = ManifoldCorrectionTree(m)
plot_comparison(mc, X_init, dims, m.data_labels, m.data_ulabels, 0)
return m
def example_deng(optimize=True, plot=True):
import pandas as pd, os
import GPy, numpy as np
from topslam.filtering import filter_RNASeq
# Reproduceability, BGPLVM has local optima
np.random.seed(42)
# This is the process of how we loaded the data:
ulabels = [
'Zygote',
'2-cell embryo',
'Early 2-cell blastomere',
'Mid 2-cell blastomere',
'Late 2-cell blastomere',
'4-cell blastomere',
'8-cell blastomere',
'16-cell blastomere',
'Early blastocyst cell',
'Mid blastocyst cell',
'Late blastocyst cell',
'fibroblast',
'adult liver',
]
folder_path = os.path.expanduser('~/tmp/Deng')
csv_file = os.path.join(folder_path, 'filtered_expression_values.csv')
if os.path.exists(csv_file):
print('Loading previous filtered data: {}'.format(csv_file))
Y_bgplvm = pd.read_csv(csv_file, index_col=[0, 1, 2], header=0)
else:
print('Loading data:')
data = GPy.util.datasets.singlecell_rna_seq_deng()
if not os.path.exists(folder_path):
os.mkdir(folder_path)
Ydata = data['Y'].copy()
Ydata.columns = Ydata.columns.to_series().apply(str.upper)
Ydata = Ydata.reset_index().set_index('index', append=True)
Ydata['labels'] = data['labels'].values
Ydata = Ydata.set_index('labels', append=True)
Ydata = Ydata.reorder_levels([0, 2, 1])
Ydata = Ydata.reset_index([0, 2]).loc[ulabels].set_index(
['level_0', 'index'], append=True)
Y = Ydata.copy()
Y.columns = [c.split('.')[0] for c in Y.columns]
Y_bgplvm = filter_RNASeq(Y)
print('\nSaving data to tmp file: {}'.format(csv_file))
Y_bgplvm.to_csv(csv_file)
labels = Y_bgplvm.index.get_level_values(0).values
Ymean = Y_bgplvm.values.mean()
Ystd = Y_bgplvm.values.std()
Y_m = Y_bgplvm.values
Y_m -= Ymean
Y_m /= Ystd
# get the labels right for split experiments
# get the labels right for 8 and split
new_8_labels = []
for _l in Y_bgplvm.loc['8-cell blastomere'].index.get_level_values(1):
_l = _l.split('-')[0]
if not ('split' in _l):
new_8_labels.append('8')
elif not ('pooled' in _l):
new_8_labels.append('8 split')
else:
new_8_labels.append('8 split')
labels[labels == '8-cell blastomere'] = new_8_labels
# get the labels right for 16 and split
new_16_labels = []
for _l in Y_bgplvm.loc['16-cell blastomere'].index.get_level_values(1):
_l = _l.split('-')[0]
if not ('split' in _l):
new_16_labels.append('16')
elif not ('pooled' in _l):
new_16_labels.append('16 split')
else:
new_16_labels.append('16 split')
labels[labels == '16-cell blastomere'] = new_16_labels
ulabels = []
for lab in labels:
if lab not in ulabels:
ulabels.append(lab)
short_labels = labels.copy()
_ulabels_convert = np.array([
'Z', # Z',
'E', # Em',
'2', # Bm E',
'2', # Bm M',
'2', # Bm L',
'4',
'8',
'8 s',
'16',
'16 s',
'Bz', # E',
'Bz', # M',
'Bz', # L'
'F',
'L'
])
short_ulabels = []
for lab, nlab in zip(ulabels, _ulabels_convert):
short_labels[short_labels == lab] = nlab
if nlab not in short_ulabels:
short_ulabels.append(nlab)
from topslam.optimization import run_methods, methods, create_model, optimize_model
X_init, dims = run_methods(Y_m, methods)
m = create_model(Y_m, X_init, num_inducing=25)
m.Ymean = Ymean
m.Ystd = Ystd
m.data_labels = short_labels
m.data_ulabels = short_ulabels
m.data = Y_bgplvm
m.X_init = X_init
m.dims = dims
if optimize:
optimize_model(m)
if plot:
mc = ManifoldCorrectionTree(m)
plot_comparison(mc, X_init, dims, m.data_labels, m.data_ulabels, 0)
return m
