def find_alignments_table(datasets, knn=KNN, approx=APPROX, verbose=VERBOSE,
prenormalized=False, geosketch=False,
geosketch_max=20000):
if not prenormalized:
datasets = [ normalize(ds, axis=1) for ds in datasets ]
if geosketch:
# Only match cells in geometric sketches.
from ample import gs, uniform
global gs_idxs
if gs_idxs is None:
gs_idxs = [ uniform(X, geosketch_max, replace=False)
if X.shape[0] > geosketch_max else range(X.shape[0])
for X in datasets ]
datasets = [ datasets[i][gs_idx, :] for i, gs_idx in enumerate(gs_idxs) ]
table = {}
for i in range(len(datasets)):
if len(datasets[:i]) > 0:
fill_table(table, i, datasets[i], datasets[:i], knn=knn,
approx=approx)
if len(datasets[i+1:]) > 0:
fill_table(table, i, datasets[i], datasets[i+1:],
knn=knn, base_ds=i+1, approx=approx)
# Count all mutual nearest neighbors between datasets.
matches = {}
table1 = {}
if verbose > 1:
table_print = np.zeros((len(datasets), len(datasets)))
for i in range(len(datasets)):
for j in range(len(datasets)):
if i >= j:
continue
if not (i, j) in table or not (j, i) in table:
continue
match_ij = table[(i, j)]
match_ji = set([ (b, a) for a, b in table[(j, i)] ])
matches[(i, j)] = match_ij & match_ji
table1[(i, j)] = (max(
float(len(set([ idx for idx, _ in matches[(i, j)] ]))) /
datasets[i].shape[0],
float(len(set([ idx for _, idx in matches[(i, j)] ]))) /
datasets[j].shape[0]
))
if verbose > 1:
table_print[i, j] += table1[(i, j)]
if geosketch:
# Translate matches within geometric sketches to original indices.
matches_mnn = matches[(i, j)]
matches[(i, j)] = [
(gs_idxs[i][a], gs_idxs[j][b]) for a, b in matches_mnn
]
if verbose > 1:
print(table_print)
return table1, table_print, matches
else:
return table1, None, matches
def integrate_sketch(datasets_dimred,
integration_fn,
integration_fn_args={},
sampling_fn=gs,
N=2000,
n_iter=1):
sketch_idxs = [
sorted(
set(gs(X, N, replace=False)) | set(uniform(X, N, replace=False)))
for X in datasets_dimred
]
datasets_sketch = [X[idx] for X, idx in zip(datasets_dimred, sketch_idxs)]
for _ in range(n_iter):
datasets_int = integration_fn(datasets_sketch[:],
**integration_fn_args)
labels = []
curr_label = 0
for i, a in enumerate(datasets_sketch):
labels += list(np.zeros(a.shape[0]) + curr_label)
curr_label += 1
labels = np.array(labels, dtype=int)
for i, (X_dimred,
X_sketch) in enumerate(zip(datasets_dimred, datasets_sketch)):
X_int = datasets_int[i]
neigh = NearestNeighbors(n_neighbors=3).fit(X_dimred)
_, neigh_idx = neigh.kneighbors(X_sketch)
ds_idxs, ref_idxs = [], []
for ref_idx in range(neigh_idx.shape[0]):
for k_idx in range(neigh_idx.shape[1]):
ds_idxs.append(neigh_idx[ref_idx, k_idx])
ref_idxs.append(ref_idx)
bias = transform(X_dimred,
X_int,
ds_idxs,
ref_idxs,
15,
batch_size=1000)
datasets_int[i] = X_dimred + bias
return datasets_int
labels += list(np.zeros(X_dimred.shape[0]) + i)
X_dimred = np.concatenate(Xs)
cell_labels = np.array(labels, dtype=int)
expected = np.array([1., 1. / 10, 1. / 100])
expected = np.array(expected) / sum(expected)
print(expected)
from ample import gs, gs_gap, srs, uniform
samp_idx = gs_gap(X_dimred, 3000, replace=True)
report_cluster_counts(cell_labels[samp_idx])
print('')
samp_idx = srs(X_dimred, 3000, replace=True)
report_cluster_counts(cell_labels[samp_idx])
print('')
samp_idx = uniform(X_dimred, 3000, replace=True)
report_cluster_counts(cell_labels[samp_idx])
exit()
experiments(
X_dimred,
NAMESPACE,
cell_labels=cell_labels,
#rare=True, rare_label=2,
#entropy=True,
kl_divergence=True,
expected=expected,
#max_min_dist=True
)
from ample import gs, gs_gap, uniform
gs_idx = gs(X_dimred, 110, replace=False)
write_table(X[gs_idx, :].toarray(), genes, 'data/pseudotime/' + NAMESPACE + '_gs')
report_cluster_counts(cell_labels[gs_idx])
with open('data/pseudotime/mono_macro_meta_gs.txt', 'w') as of:
of.write('Label\n')
i = 0
for idx in range(X.shape[0]):
if idx not in gs_idx:
continue
of.write('mono_macro_gs{}\t{}\n'.format(i, cell_names[idx]))
i += 1
uni_idx = uniform(X_dimred, 110, replace=False)
write_table(X[uni_idx, :].toarray(), genes, 'data/pseudotime/' + NAMESPACE + '_uni')
report_cluster_counts(cell_labels[uni_idx])
with open('data/pseudotime/mono_macro_meta_uni.txt', 'w') as of:
of.write('Label\n')
i = 0
for idx in range(X.shape[0]):
if idx not in uni_idx:
continue
of.write('mono_macro_uni{}\t{}\n'.format(i, cell_names[idx]))
i += 1
with open('data/pseudotime/mono_macro_genes.txt', 'w') as of:
of.write('gene_short_name\n')
for gene in genes:
