2dShapeSpacePortable/shapemodes.py at main · CellProfiling/2dShapeSpacePortable · 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
import os
import numpy as np
from sklearn.decomposition import PCA
import pandas as pd
import json

from lib import plotting


def shapemodes(config, fftcoeffs_path):
    lines = dict()
    df_fft = pd.read_csv(fftcoeffs_path)
    for index, row in df_fft.iterrows():
        if row.isna().any():
            config["log"].info("-- Dismissing " + row["image"] + " image due nan values")
            continue
        elif row["cell_area"] / row["nuc_area"] > config["dismiss_ratio"]:
            config["log"].info("-- Dismissing " + row["image"] + " image due to cell/nuc ratio")
            continue
        else:
            lines[row["image"]] = row[10:]

    df = pd.DataFrame(lines).transpose()
    df = df.map(lambda s: complex(s.replace("i", "j")))

    df_ = pd.concat(
        [
            pd.DataFrame(df.to_numpy().real),
            pd.DataFrame(df.to_numpy().imag),
        ],
        axis=1,
    )
    pca = PCA()  # IncrementalPCA(whiten=True) #PCA()
    pca.fit(df_)
    plotting.display_scree_plot(pca, save_dir=os.path.join(config["output_dir"], "shapemode"))

    scree = pca.explained_variance_ratio_ * 100
    for percent in np.arange(70, 100, 5):
        n_pc = np.sum(scree.cumsum() < percent) + 1
        config["log"].info(f"{n_pc} to explain {percent} % variance")
    n_pc = np.sum(scree.cumsum() < 95) + 1
    n_pc = 8 if n_pc < 8 else n_pc
    pc_names = [f"PC{c}" for c in range(1, 1 + len(pca.components_))]
    pc_keep = [f"PC{c}" for c in range(1, 1 + n_pc)]
    # pc_keep = [f"PC{c}" for c in range(1, 1 + 6)]
    matrix_of_features_transform = pca.transform(df_)
    df_trans = pd.DataFrame(data=matrix_of_features_transform.copy())
    df_trans.columns = pc_names
    df_trans.index = df.index
    df_trans[list(set(pc_names) - set(pc_keep))] = 0
    df_trans.to_csv(os.path.join(config["output_dir"], "shapemode", "transformed_matrix.csv"))

    # Cell density on major PC
    plotting.plot_pc_density(df_trans["PC1"], df_trans["PC2"], save_path=os.path.join(config["output_dir"], "shapemode", "PC1vsPC2_cell_density.png"))
    plotting.plot_pc_density(df_trans["PC2"], df_trans["PC3"], save_path=os.path.join(config["output_dir"], "shapemode", "PC2vsPC3_cell_density.png"))

    pm = plotting.PlotShapeModes(
        pca,
        df_trans,
        config['n_coeffs'],
        pc_keep,
        scaler=None,
        complex_type=False
    )
    pm.plot_avg_cell(dark=False, save_dir=os.path.join(config["output_dir"], "shapemode"))

    n_ = min(10, len(df_trans))  # number of random cells to plot
    cells_assigned = dict()
    for pc in pc_keep:
        pm.plot_shape_variation_gif(pc, dark=False, save_dir=os.path.join(config["output_dir"], "shapemode"))
        pm.plot_shape_variation(pc, dark=False, save_dir=os.path.join(config["output_dir"], "shapemode"))
        pm.plot_pc_hist(pc, save_dir=os.path.join(config["output_dir"], "shapemode"))

        pc_indexes_assigned, bin_links = pm.assign_cells(pc)
        cells_assigned[pc] = [list(b) for b in bin_links]

        plotting.plot_example_cells(bin_links,
                                    n_coef=config['n_coeffs'],
                                    cells_per_bin=5,
                                    shape_coef_path=fftcoeffs_path,
                                    save_path=os.path.join(config["output_dir"], "shapemode", pc + "_example_cells.png"))

    with open(os.path.join(config["output_dir"], "shapemode", "cells_assigned_to_pc_bins.json"), "w") as fp:
        json.dump(cells_assigned, fp)