Evaluate Pearson correlation score and cell type accuracy score

import pandas as pd
import numpy as np
import scanpy as sc
import anndata as ad
import umap
import os
import matplotlib.pyplot as plt
import matplotlib as mpl
from matplotlib.cm import get_cmap
from matplotlib.colors import ListedColormap
from matplotlib.lines import Line2D

from sklearn.neighbors import NearestNeighbors
import scipy.stats

Load results

res_path = "Results/INSPIRE_registration_starmapplus"
adata_0_ours = sc.read_h5ad(res_path + "/adata_st_list_0_loc_aligned.h5ad")
adata_1_ours = sc.read_h5ad(res_path + "/adata_st_list_1_loc_aligned.h5ad")

print("Load data 1...")

data_dir = "3d-alignment/STARmapPLUS"
adata_1 = sc.read_h5ad(data_dir + "/adata_8m_disease_rep1.h5ad")
adata_1.var_names_make_unique()

print("Load data 2...")

data_dir = "3d-alignment/STARmapPLUS"
adata_2 = sc.read_h5ad(data_dir + "/adata_8m_disease_rep2.h5ad")
adata_2.var_names_make_unique()

Load data 1...
Load data 2...

adata_st_list = [adata_1, adata_2]

adata_0 = adata_st_list[0]
adata_0.obs.index = adata_0.obs.index + "-1"

adata_1 = adata_st_list[1]
adata_1.obs.index = adata_1.obs.index + "-0"

adata_0.obsm["loc_ours"] = adata_1_ours[adata_0.obs.index, :].obsm["loc_aligned"]
adata_1.obsm["loc_ours"] = adata_0_ours[adata_1.obs.index, :].obsm["loc_aligned"]
adata_0.obs["celltype"] = adata_1_ours[adata_0.obs.index, :].obs["top_level_cell_type"]
adata_1.obs["celltype"] = adata_0_ours[adata_1.obs.index, :].obs["top_level_cell_type"]

s = 1
f = plt.figure(figsize=(2.45, 2.45))

ax2 = f.add_subplot(1,1,1)
ax2.scatter(adata_1.obsm["loc_ours"][:,0], -adata_1.obsm["loc_ours"][:,1], facecolors="tab:orange", s=s, rasterized=True, marker='o', edgecolors='none')
ax2.scatter(adata_0.obsm["loc_ours"][:,0], -adata_0.obsm["loc_ours"][:,1], facecolors="tab:blue", s=s, rasterized=True, marker='o', edgecolors='none')
ax2.axis('equal')
ax2.tick_params(axis='both',bottom=False, top=False, left=False, right=False, labelleft=False, labelbottom=False, grid_alpha=0)

adata_0.obsm["spatial"] = adata_0.obsm["loc_ours"]
sc.pl.spatial(adata_0, color="celltype", spot_size=100.)

adata_1.obsm["spatial"] = adata_1.obsm["loc_ours"]
sc.pl.spatial(adata_1, color="celltype", spot_size=100.)

Calculate the scores

k = 6

neigh = NearestNeighbors(n_neighbors=k, metric="euclidean")
neigh.fit(adata_1.obsm["loc_ours"])
NN = neigh.kneighbors(adata_0.obsm["loc_ours"], k, return_distance=False)

PCC = []
for i in range(NN.shape[0]):
    x0 = adata_0.X[i, :]
    x1_mat = adata_1.X[NN[i, :], :]
    pcc_cell = np.corrcoef(x0.reshape((1,-1)), x1_mat)[0, :][1:]
    PCC.append(np.max(pcc_cell))

print("Pearson correlation score =", np.mean(PCC))

Pearson correlation score = 0.2562931829758295

neigh = NearestNeighbors(n_neighbors=k, metric="euclidean")
neigh.fit(adata_1.obsm["loc_ours"])
NN = neigh.kneighbors(adata_0.obsm["loc_ours"], k, return_distance=False)

acc = []
for i in range(NN.shape[0]):
    ct_0 = adata_0.obs["celltype"][i]
    ct_1_nn = adata_1.obs["celltype"][NN[i, :]]
    s_same = np.sum(ct_1_nn == ct_0)
    if s_same == 0:
        acc.append(0)
    else:
        acc.append(1)

print("Cell type accuracy score =", np.mean(acc))

Cell type accuracy score = 0.7369899828976301