subcell-analysis/harmonize.py at main · CellProfiling/subcell-analysis · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
import os
import pandas as pd
import torch
from harmony import harmonize
from utils.feature_analysis import get_per_gene_mean_features

if __name__ == "__main__":
    exp_name_dir_dict = {
        "ViT-ProtS-Pool": "/scratch/groups/emmalu/subcell_ankit/features/ViT-ProtS-Pool/hpa_features/all_features.pth",
        "MAE-CellS-ProtS-Pool": "/scratch/groups/emmalu/subcell_ankit/features/MAE-CellS-ProtS-Pool/hpa_features/all_features.pth",
        "DINO4Cells-HPA": "/scratch/groups/emmalu/subcell_ankit/features/DINO4Cells-HPA/hpa_features/all_features.pth",
        "ViT-Supervised": "/scratch/groups/emmalu/subcell_ankit/features/ViT-Supervised/hpa_features/all_features.pth",
        "bestfitting": "/scratch/groups/emmalu/subcell_ankit/features/bestfitting/hpa_features/all_features.pth",
    }

    harmonize_cats = [
        ["if_plate_id"],
        ["microscope"],
        ["atlas_name"],
        ["if_plate_id", "microscope"],
        ["if_plate_id", "microscope", "atlas_name"],
    ]
    cell_lines = ["U2OS", "all"]

    meta_df = (
        pd.read_csv("annotations/IF-image-orig-bboxes.csv")
        .groupby(["if_plate_id", "position", "sample", "image_width"])
        .first()
        .reset_index()
    )
    meta_df["microscope"] = meta_df["image_width"].apply(
        lambda x: "M1" if x >= 2048 else "M2"
    )

    all_path_mean_file_dict = {}

    for exp_name, dir_path in exp_name_dir_dict.items():
        method_save_folder = f"{os.path.dirname(dir_path)}"

        df, feature_data = torch.load(dir_path, map_location="cpu", weights_only=False)

        for cell_line in cell_lines:
            if os.path.exists(f"{method_save_folder}/mean_feat_{cell_line}.csv"):
                print(
                    f"Mean features for {exp_name} with cell line {cell_line} already exist."
                )
            else:
                feat_dict = get_per_gene_mean_features(
                    df,
                    feature_data,
                    cell_lines=cell_line if cell_line != "all" else None,
                )
                feat_df = pd.DataFrame(feat_dict).T
                feat_df.to_csv(f"{method_save_folder}/mean_feat_{cell_line}.csv")

            all_path_mean_file_dict[f"{exp_name}_{cell_line}"] = (
                f"{method_save_folder}/mean_feat_{cell_line}.csv"
            )

        df1 = pd.merge(
            df,
            meta_df[["if_plate_id", "position", "sample", "microscope"]],
            on=["if_plate_id", "position", "sample"],
            how="left",
        )
        df1["microscope"] = df1["microscope"].fillna("M1")

        if exp_name in ["bestfitting", "ViT-Supervised"]:
            continue

        for harmonize_cat in harmonize_cats:
            if os.path.exists(
                f"{method_save_folder}/harmonized_features_{'_'.join(harmonize_cat)}.pth"
            ):
                print(
                    f"Harmonized features for {exp_name} with categories {harmonize_cat} already exist."
                )
                df, harmonized_features = torch.load(
                    f"{method_save_folder}/harmonized_features_{'_'.join(harmonize_cat)}.pth",
                    map_location="cpu",
                    weights_only=False,
                )
            else:
                print(f"Harmonizing {exp_name} with categories {harmonize_cat}")
                harmonized_features = harmonize(
                    X=feature_data.numpy(),
                    batch_mat=df1,
                    batch_key=harmonize_cat,
                    use_gpu=True,
                    verbose=True,
                    random_state=42,
                )
                harmonized_features = torch.tensor(
                    harmonized_features, dtype=torch.float32
                )

                torch.save(
                    (df, harmonized_features),
                    f"{method_save_folder}/harmonized_features_{'_'.join(harmonize_cat)}.pth",
                )

            for cell_line in cell_lines:
                if os.path.exists(
                    f"{method_save_folder}/mean_feat_{cell_line}_harmonized_{'_'.join(harmonize_cat)}.csv"
                ):
                    print(
                        f"Mean features for {exp_name} with cell line {cell_line} and harmonization {harmonize_cat} already exist."
                    )
                else:
                    feat_dict = get_per_gene_mean_features(
                        df,
                        harmonized_features,
                        cell_lines=cell_line if cell_line != "all" else None,
                    )
                    feat_df = pd.DataFrame(feat_dict).T
                    feat_df.to_csv(
                        f"{method_save_folder}/mean_feat_{cell_line}_harmonized_{'_'.join(harmonize_cat)}.csv",
                    )
                all_path_mean_file_dict[
                    f"{exp_name}_{cell_line}_harmonized_{'_'.join(harmonize_cat)}"
                ] = f"{method_save_folder}/mean_feat_{cell_line}_harmonized_{'_'.join(harmonize_cat)}.csv"

    all_path_mean_file_df = pd.DataFrame.from_dict(
        all_path_mean_file_dict, orient="index", columns=["path"]
    )
    all_path_mean_file_df.index.name = "experiment_cell_line"
    all_path_mean_file_df.to_csv("annotations/mean_features_paths.csv", index=True)