def NMFreg(
counts,
coords,
size,
metacell_dict,
gene_intersection,
num_atlas_clusters,
celltype_to_factor_dict,
celltype_dict,
plot_size_dict,
):
puckcounts = counts[["barcode"] + gene_intersection]
puckcounts = puckcounts.set_index(counts["barcode"])
puckcounts = puckcounts.drop("barcode", axis=1)
cell_totalUMI = np.sum(puckcounts, axis=1)
puckcounts_cellnorm = np.divide(puckcounts, cell_totalUMI[:, None])
puckcounts_scaled = StandardScaler(
with_mean=False).fit_transform(puckcounts_cellnorm)
XsT = puckcounts_scaled.T
Hs_hat = []
for b in tqdm(range(XsT.shape[1])):
h_hat = scipy.optimize.nnls(WaT, XsT[:, b])[0]
if b == 0:
Hs_hat = h_hat
else:
Hs_hat = np.vstack((Hs_hat, h_hat))
Hs = pd.DataFrame(Hs_hat)
Hs["barcode"] = puckcounts.index.tolist()
Hs_norm = StandardScaler(with_mean=False).fit_transform(
Hs.drop("barcode", axis=1))
Hs_norm = pd.DataFrame(Hs_norm)
Hs_norm["barcode"] = puckcounts.index.tolist()
maxloc_s = Hs_norm.drop("barcode", axis=1).values.argmax(axis=1)
barcode_clusters = pd.DataFrame()
barcode_clusters["barcode"] = Hs_norm["barcode"]
barcode_clusters["max_factor"] = maxloc_s
barcode_clusters["atlas_cluster"] = barcode_clusters["barcode"]
for c in range(1, num_atlas_clusters + 1):
condition = np.isin(barcode_clusters["max_factor"],
celltype_to_factor_dict[c])
barcode_clusters["atlas_cluster"][condition] = c
bead_deconv_df = Hs_norm.apply(
lambda x: deconv_factor_to_celltype(
row=x,
adict=factor_to_celltype_dict,
K=K,
num_atlas_clusters=num_atlas_clusters,
),
axis=1,
)
bead_deconv_df.insert(0, "barcode", Hs_norm["barcode"])
bead_deconv_df.columns = ["barcode"
] + (bead_deconv_df.columns[1:] + 1).tolist()
bead_deconv_df = pd.DataFrame(bead_deconv_df)
bead_deconv_df = bead_deconv_df.rename(columns=celltype_dict)
maxloc_ct = bead_deconv_df.drop("barcode",
axis=1).values.argmax(axis=1) + 1
bead_maxct_df = pd.DataFrame()
bead_maxct_df["barcode"] = bead_deconv_df["barcode"]
bead_maxct_df["max_cell_type"] = maxloc_ct
return Hs, Hs_norm, puckcounts, bead_deconv_df, barcode_clusters, bead_maxct_df
