scaled = counts.divide(counts.sum(axis=0), axis=1) * 10000
gene_means = scaled.mean(axis=1)
valid_genes = gene_means.index[gene_means < highXMeanCutoff]
hs_results = hs_results.loc[valid_genes & hs_results.index]
if topN is None:
hs_genes = hs_results.index[hs_results.FDR < fdrThresh]
else:
hs_genes = hs_results.sort_values('Z').tail(topN).index
hs_genes = hs_genes & counts.index
counts = counts.loc[hs_genes]
c_counts = local_stats_pairs.create_centered_counts(counts.values, model,
num_umi.values)
# Run the process
eg2s = np.array([
conditional_eg2(c_counts[i], hs.neighbors.values, hs.weights.values)
for i in range(c_counts.shape[0])
])
def _map_fun_parallel_pairs_centered(rowpair):
global g_neighbors
global g_weights
global g_counts
global g_eg2s
return _compute_hs_pairs_inner_centered_cond_sym_shuff(
hs = hotspot.Hotspot(counts, model='danb', latent=latent, umi_counts=num_umi)
hs.create_knn_graph(weighted_graph=weighted_graph,
n_neighbors=n_neighbors,
neighborhood_factor=3)
neighbors = hs.neighbors.values
weights = hs.weights.values
norm_counts = counts.values / num_umi.values.reshape((1, -1)) * 10000
norm_counts = np.log(norm_counts + 1)
from hotspot.local_stats_pairs import create_centered_counts
norm_counts = create_centered_counts(counts.values, hs.model, num_umi.values)
geary_results = []
moran_results = []
for i in tqdm(range(norm_counts.shape[0])):
exp_vals = norm_counts[i, :]
g = geary(exp_vals, neighbors, weights)
geary_results.append(g)
m = moran(exp_vals, neighbors, weights)
moran_results.append(m)
results = pd.DataFrame({
'GearyC': geary_results,
counts = ds[:, :]
scaled = ds.layers['scaled'][:, :]
gene_info = ds.ra['EnsID', 'Symbol']
num_umi = ds.ca['NumUmi'][:]
gene_info = pd.DataFrame(
gene_info, columns=['EnsID', 'Symbol']).set_index('EnsID')
counts = pd.DataFrame(counts, index=gene_info.index, columns=barcodes)
scaled = pd.DataFrame(scaled, index=gene_info.index, columns=barcodes)
num_umi = pd.Series(num_umi, index=barcodes)
counts_hs = counts.loc[hs_genes]
scaled_hs = scaled.loc[hs_genes]
from hotspot.local_stats_pairs import create_centered_counts
model = 'bernoulli'
counts_hs_centered = create_centered_counts(counts_hs.values, model, num_umi.values)
counts_hs_centered = pd.DataFrame(
counts_hs_centered, index=counts_hs.index, columns=counts_hs.columns)
# %% SVD
from sklearn.decomposition import TruncatedSVD
model = TruncatedSVD(n_components=5)
results = model.fit_transform(counts_hs_centered.values.T)
results.shape
# %% TSNE
from sklearn.manifold import TSNE