def create_embedding_datamatrix(dm, n_components):
em = dataclasses.datamatrix(
rowname=dm.rowname,
rowlabels=dm.rowlabels.copy(),
rowmeta=dm.rowmeta.copy(),
columnname='latent_component',
columnlabels=np.array(['LC' + str(x) for x in range(n_components)],
dtype='object'),
columnmeta={},
matrixname='sdae_embedding_of_' + dm.matrixname,
matrix=np.zeros((dm.shape[0], n_components), dtype='float32'))
return em
with open('aligned_matrices_{0}/gene_atb_{1}.pickle'.format(
analysis_version, datasetabbrev),
mode='rb') as fr:
gene_atb = pickle.load(fr)
with open('aligned_matrices_{0}/gene_cst_{1}.pickle'.format(
analysis_version, datasetabbrev),
mode='rb') as fr:
gene_cst = pickle.load(fr)
# initialize outputs
atb_cst = {}
atb_cst['test_statistic_values'] = dc.datamatrix(
rowname=gene_atb.columnname,
rowlabels=gene_atb.columnlabels.copy(),
rowmeta=copy.deepcopy(gene_atb.columnmeta),
columnname=gene_cst.columnname,
columnlabels=gene_cst.columnlabels.copy(),
columnmeta=copy.deepcopy(gene_cst.columnmeta),
matrixname='atb_cluster_correlation',
matrix=np.zeros((gene_atb.shape[1], gene_cst.shape[1]), dtype='float64'))
atb_cst['pvalues'] = copy.deepcopy(atb_cst['test_statistic_values'])
# computation
starttime = time.time()
print('starting {0!s} permutations...'.format(numperm))
atb_cst['test_statistic_values'].matrix, atb_cst[
'pvalues'].matrix = feature_selection_test_function(X=gene_cst.matrix,
Y=gene_atb.matrix,
numperm=numperm)
atb_cst['pvalues'].matrix = atb_cst['pvalues'].matrix.T
if batch == 0:
sys.path.append(custompath)
del custompath, custompaths
import numpy as np
import machinelearning.dataclasses as dc
import pickle
# load clusters
with open('clusters.pickle', 'rb') as fr:
gene_syms, gene_ids, cluster_ids = pickle.load(fr)
unique_cluster_ids = np.array([str(x) for x in np.unique(cluster_ids)],
dtype='object')
# create matrix
gene_clust = dc.datamatrix(
rowname='gene_sym',
rowlabels=gene_syms,
rowmeta={'gene_id': gene_ids},
columnname='cluster_id',
columnlabels=unique_cluster_ids,
columnmeta={},
matrixname=
'gene_cluster_assignments_from_denoising_autoencoder_applied_to_GTEX',
matrix=np.zeros((gene_syms.size, unique_cluster_ids.size), dtype='bool'))
for j, cluster_id in enumerate(gene_clust.columnlabels):
gene_clust.matrix[:, j] = cluster_ids == int(cluster_id)
# write matrix
with open('gene_cluster_matrix.pickle', 'wb') as fw:
pickle.dump(gene_clust, fw)
) / np.pi # divide by pi if similarity scores can be negative, otherwise divide by pi/2
else:
D = np.arccos(D) / (np.pi / 2)
return D
else:
raise ValueError('invalid distance metric')
gene_atb = dc.datamatrix(rowname='GeneSym',
rowlabels=np.concatenate(
(train.rowlabels, valid.rowlabels,
test.rowlabels)),
rowmeta={
x: np.concatenate(
(train.rowmeta[x], valid.rowmeta[x],
test.rowmeta[x]))
for x in train.rowmeta
},
columnname='Tissue',
columnlabels=train.columnlabels.copy(),
columnmeta={},
matrixname='zscored_tissue_expression',
matrix=np.concatenate(
(train.matrix, valid.matrix, test.matrix), 0))
gene_proj = copy.deepcopy(gene_atb)
gene_proj.columnlabels = np.array(['X', 'Y'], dtype='object')
gene_proj.columnname = 'Neuron'
gene_proj.matrixname = '2d_dnn_projection_of_zscored_tissue_expression'
gene_proj.matrix = sess.run(h[-1], feed_dict={x: gene_atb.matrix})
gene_proj.updatesizeattribute()
gene_proj.updateshapeattribute()