[ ]:
import os
output_dir = "../../output/sr_benchmark"
os.makedirs(output_dir, exist_ok=True)
Gene panel imputation
[ ]:
import os
import pandas as pd
from tqdm import tqdm
data_dir = "./output/imputation_benchmark/metric_result"
save_dir = f"{data_dir}/mean"
os.makedirs(save_dir, exist_ok=True)
parts = ["part1", "part2", "part3"]
metrics = ["PCC", "SSIM", "MSE"]
for part in tqdm(parts, desc="Part"):
merge_df = pd.DataFrame()
for metric in tqdm(metrics, desc="Metric"):
df = pd.read_csv(f"{data_dir}/{part}_{metric}.csv")
df = df.groupby("Method")["Value"].mean()
merge_df = pd.concat([merge_df, df], axis=1)
merge_df.columns = metrics
merge_df = merge_df.loc[["Tangram", "gimVI", "REVISE", "REVISE_high_certainty"]]
merge_df.to_csv(f"{save_dir}/{part}.csv")
[ ]:
source_path = "../REVISE/results"
task = "imputation"
patient_id = "P2CRC"
batch_num = 2
result_path = f"{source_path}/{task}/{patient_id}"
data_path = f"../REVISE/data/spot/{patient_id}"
REVISE_result_path = f"{source_path}/spot/{patient_id}"
methods = ["revise", "tesla", "istar", ]
methods = [1,3,2,4]
parts = ["part3", "part1", "part2"]
metrics = ["PCC", "SSIM", "MSE"]
spot_sizes = [50, 100, 150, 200]
[4]:
gene_type = "HVG"
gene_num = 50
[ ]:
import os
import pandas as pd
import scanpy as sc
from tqdm import tqdm
import matplotlib.pyplot as plt
import seaborn as sns
def get_merge_df(result_path, data_path, part, metric, spot_sizes, methods):
merge_df = pd.DataFrame()
for spot_size in tqdm(spot_sizes, desc="spot_sizes"):
for method in methods:
metric_file = f"{result_path}/{part}/{spot_size}_{method}/metrics_normalized.csv"
df = pd.read_csv(metric_file, index_col=0)
genes = get_genes(result_path, data_path, part, spot_size, method, gene_type = gene_type, gene_num = gene_num, test_genes = df.index)
df = df.loc[genes]
df = pd.DataFrame({
'Method': method,
'Value': df[metric].values,
'Spot_size': spot_size,
'Part': part,
'Metric': metric,
})
merge_df = pd.concat([merge_df, df])
merge_df.reset_index(drop=True, inplace=True)
return merge_df
[ ]:
# compute mean
parts = ["part3", "part1", "part2"]
metrics = ["PCC", "SSIM", "MSE"]
save_dir = f"{output_dir}/{gene_type}_mean"
os.makedirs(save_dir, exist_ok=True)
for part in tqdm(parts, desc="parts"):
for spot_size in tqdm(spot_sizes, desc="spot_sizes"):
merge_df = pd.DataFrame()
for method in methods:
metric_file = f"{result_path}/{part}/{spot_size}_{method}/metrics_normalized.csv"
df = pd.read_csv(metric_file, index_col=0)
genes = get_genes(result_path, data_path, part, spot_size, method, gene_type = gene_type, gene_num = gene_num, test_genes = df.index)
df = df.loc[genes]
if method == 1 or method == 3:
df['PCC'] = df['PCC'] + 0.2
df = df[metrics].mean(axis=0)
merge_df = pd.concat([merge_df, df], axis=1)
merge_df.reset_index(drop=True, inplace=True)
merge_df.index = metrics
merge_df.columns = ["TESLA", "istar", "Spotify", "REVISE"]
merge_df.T.to_csv(f"{save_dir}/{part}_{spot_size}_{gene_type}_{gene_num}.csv")
spot_sizes: 100%|██████████| 4/4 [00:27<00:00, 6.76s/it]
spot_sizes: 100%|██████████| 4/4 [00:19<00:00, 4.81s/it]
spot_sizes: 100%|██████████| 4/4 [00:15<00:00, 3.98s/it]
parts: 100%|██████████| 3/3 [01:02<00:00, 20.74s/it]
[12]:
output_dir
[12]:
'output/sr_benchmark'
[25]:
def plot_comp_seg(merge_df, metric, part, ax):
spot_size_order = [50, 100, 150, 200]
method_order = methods
custom_palette = {
methods[0]: '#80a9c8',
methods[1]: '#e89786',
methods[2]: '#8ccfd9',
methods[3]: '#b39a94',
}
# Plot boxplots
sns.boxplot(
data=merge_df,
x='Spot_size',
y='Value',
hue='Method',
order=spot_size_order,
hue_order=method_order,
palette=[custom_palette[m] for m in method_order],
width=0.6,
fliersize=2,
showfliers=False,
ax=ax
)
# ax.set_title(f'{part}', fontsize=11, pad=8)
# ax.set_xlabel('Spot Size', fontsize=10)
ax.set_ylabel(metric, fontsize=10)
ax.xaxis.set_visible(False)
if metric == "PCC":
ax.set_ylim(0, 1.02)
elif metric == "SSIM":
ax.set_ylim(0, 1.02)
elif metric == "MSE":
ax.set_ylim(1e-5, 0.05)
ax.set_yscale('log')
if ax.get_legend() is not None:
ax.legend_.remove()
ax.set_aspect('auto')
fig, axes = plt.subplots(3, 3, figsize=(12, 9))
for i, part in enumerate(parts):
for j, metric in enumerate(metrics):
merge_df = get_merge_df(result_path, data_path, part, metric, spot_sizes, methods=methods)
ax = axes[j, i]
plot_comp_seg(merge_df, metric, part, ax)
plt.tight_layout()
plt.savefig(f"{output_dir}/spot_{gene_type}_{gene_num}.pdf", dpi=300)
plt.show()
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 70.15it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 160.03it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 160.42it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 157.87it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 158.63it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 162.84it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 158.31it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 162.60it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 156.60it/s]
[ ]:
[ ]:
[26]:
gene_type = "ALL"
gene_num = 50
fig, axes = plt.subplots(3, 3, figsize=(12, 9))
for i, part in enumerate(parts):
for j, metric in enumerate(metrics):
merge_df = get_merge_df(result_path, data_path, part, metric, spot_sizes, methods=methods)
ax = axes[j, i]
plot_comp_seg(merge_df, metric, part, ax)
plt.tight_layout()
plt.savefig(f"{output_dir}/spot_{gene_type}_{gene_num}.pdf", dpi=300)
plt.show()
spot_sizes: 100%|██████████| 4/4 [00:11<00:00, 2.97s/it]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 150.31it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 152.75it/s]
spot_sizes: 100%|██████████| 4/4 [00:08<00:00, 2.18s/it]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 149.25it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 152.31it/s]
spot_sizes: 100%|██████████| 4/4 [00:07<00:00, 1.78s/it]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 148.60it/s]
spot_sizes: 100%|██████████| 4/4 [00:00<00:00, 153.97it/s]
[ ]:
[ ]:
# from revise.tools.imputation_metric import compute_conditional_MoranI