clone2vec 0.1.0__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
clone2vec/datasets.py ADDED
@@ -0,0 +1,94 @@
1
+ from __future__ import annotations
2
+
3
+ from scanpy import read, AnnData
4
+ from pathlib import Path
5
+
6
+ zenodo_record_lt = "15334396"
7
+
8
+ def Weinreb_in_vitro(
9
+ file_path: str | Path = "data/Weinreb_in_vitro.h5ad",
10
+ ) -> AnnData:
11
+ """
12
+ Dataset from [PMID: 31974159] with in vitro hematopoiesis.
13
+
14
+ Parameters
15
+ ----------
16
+ file_path : str | Path, optional
17
+ Path where .h5ad-container will be stored, by default "data/Weinreb_in_vitro.h5ad".
18
+
19
+ Returns
20
+ -------
21
+ AnnData
22
+ Annotated data matrix with the dataset.
23
+ """
24
+ url = f"https://zenodo.org/records/{zenodo_record_lt}/files/Weinreb_in_vitro.h5ad"
25
+
26
+ adata = read(file_path, backup_url=url, sparse=True, cache=True)
27
+ return adata
28
+
29
+ def Erickson_murine_development(
30
+ perturbed: bool = False,
31
+ region: str = "trunk",
32
+ subset: str = "all",
33
+ file_path: str | Path | None = None,
34
+ ) -> AnnData:
35
+ """
36
+ Dataset from [PMID: 40502176] with clonal atlas of murine development.
37
+
38
+ Parameters
39
+ ----------
40
+ perturbed : bool, optional
41
+ If the dataset should contain also experiments with mosaic knockouts,
42
+ by default False.
43
+ region : str, optional
44
+ Region of interest, one of `trunk` or `head`, by default "trunk".
45
+ subset : str, optional
46
+ Subset of interest (`all`, `neurons`, `mesenchyme` or `other`), by default "all".
47
+ file_path : str | Path | None, optional
48
+ Path where .h5ad-container will be stored, by default None.
49
+
50
+ Returns
51
+ -------
52
+ AnnData
53
+ Annotated data matrix with the dataset
54
+ """
55
+ datasets = {
56
+ False: {
57
+ "trunk": {
58
+ "all": "Erickson_Trunk_Control_All.h5ad",
59
+ "mesenchyme": "Erickson_Trunk_Control_Mesenchyme.h5ad",
60
+ "neurons": "Erickson_Trunk_Control_Neurons.h5ad",
61
+ "other": "Erickson_Trunk_Control_Other.h5ad",
62
+ },
63
+ "head": {
64
+ "all": "Erickson_Head_Control_All.h5ad",
65
+ "mesenchyme": "Erickson_Head_Control_Mesenchyme.h5ad",
66
+ "neurons": "Erickson_Head_Control_Neurons.h5ad",
67
+ "other": "Erickson_Head_Control_Other.h5ad",
68
+ },
69
+ },
70
+ True: {
71
+ "trunk": {
72
+ "all": "Erickson_Trunk_Perturb_All.h5ad",
73
+ "mesenchyme": "Erickson_Trunk_Perturb_Mesenchyme.h5ad",
74
+ "neurons": "Erickson_Trunk_Perturb_Neurons.h5ad",
75
+ "other": "Erickson_Trunk_Perturb_Other.h5ad",
76
+ },
77
+ "head": {
78
+ "all": "Erickson_Head_Perturb_All.h5ad",
79
+ "mesenchyme": "Erickson_Head_Perturb_Mesenchyme.h5ad",
80
+ "neurons": "Erickson_Head_Perturb_Neurons.h5ad",
81
+ "other": "Erickson_Head_Perturb_Other.h5ad",
82
+ },
83
+ },
84
+ }
85
+
86
+ dataset = datasets[perturbed][region][subset]
87
+
88
+ if file_path is None:
89
+ file_path = f"data/{dataset}"
90
+
91
+ url = f"https://zenodo.org/records/{zenodo_record_lt}/files/{dataset}"
92
+
93
+ adata = read(file_path, backup_url=url, sparse=True, cache=True)
94
+ return adata
@@ -0,0 +1,187 @@
1
+ from __future__ import annotations
2
+
3
+ import scanpy as sc
4
+ import numpy as np
5
+ import sys
6
+
7
+ from scipy import linalg, interpolate
8
+ from contextlib import nullcontext
9
+ from tqdm import tqdm
10
+
11
+ logg = sc.logging
12
+
13
+ tw_cdf_tabular = {
14
+ -9.0: 0.000000e+00,
15
+ -8.0: 0.000000e+00,
16
+ -7.0: 0.000000e+00,
17
+ -6.0: 1.28e-09,
18
+ -5.0: 1.59e-06,
19
+ -4.0: 0.00134,
20
+ -3.5: 0.00603,
21
+ -3.0: 0.01616,
22
+ -2.5: 0.05590,
23
+ -2.0: 0.14932,
24
+ -1.5: 0.30771,
25
+ -1.0: 0.51085,
26
+ -0.5: 0.70612,
27
+ 0.0: 0.85172,
28
+ 0.5: 0.93722,
29
+ 1.0: 0.97720,
30
+ 1.5: 0.99284,
31
+ 2.0: 0.99802,
32
+ 2.5: 0.99951,
33
+ 3.0: 0.99989,
34
+ }
35
+
36
+ def _tracy_widom_cdf(x: float) -> float:
37
+ """
38
+ Computes the CDF of the Tracy-Widom distribution (with beta = 1).
39
+
40
+ Parameters
41
+ ----------
42
+ x : float
43
+ Value at which to evaluate the CDF.
44
+
45
+ Returns
46
+ -------
47
+ float
48
+ The value of the CDF at x.
49
+ """
50
+ if x > 3:
51
+ tail_prob = np.exp(-2.0/3.0 * x**1.5) / (4.0 * np.sqrt(np.pi) * x**1.5)
52
+ return 1.0 - tail_prob
53
+ elif x < -10:
54
+ return 0
55
+ else:
56
+ tw_x = np.array(list(tw_cdf_tabular.keys()))
57
+ tw_cdf = np.array(list(tw_cdf_tabular.values()))
58
+ interp = interpolate.interp1d(
59
+ tw_x,
60
+ tw_cdf,
61
+ kind="cubic",
62
+ bounds_error=False,
63
+ fill_value=(0.0, 1.0),
64
+ )
65
+ val = interp(x)
66
+ return float(np.clip(val, 0.0, 1.0))
67
+
68
+ def _calculate_statistics(X: np.ndarray) -> float:
69
+ """
70
+ Calculates the test statistic for the Tracy-Widom distribution.
71
+
72
+ Parameters
73
+ ----------
74
+ X : np.ndarray
75
+ Matrix of shape (n, p) containing the data.
76
+
77
+ Returns
78
+ -------
79
+ float
80
+ The value of the test statistic.
81
+ """
82
+ n, p = X.shape
83
+ X_centered = (X - np.mean(X, axis=0)) / np.std(X, axis=0)
84
+ n_dof = n - 1
85
+
86
+ s = linalg.svdvals(X_centered)
87
+ eigenvalues = s ** 2
88
+ lambda_1 = eigenvalues[0]
89
+ sigma_sq_est = np.mean(eigenvalues[1:]) / n_dof
90
+
91
+ mu_np = (np.sqrt(n_dof) + np.sqrt(p)) ** 2
92
+ sigma_np = (np.sqrt(n_dof) + np.sqrt(p)) * \
93
+ ((1 / np.sqrt(n_dof)) + (1 / np.sqrt(p))) ** (1/3)
94
+
95
+ stat = ((lambda_1 / sigma_sq_est) - mu_np) / sigma_np
96
+ return stat
97
+
98
+ def eigenvalue_test(
99
+ adata: sc.AnnData | np.ndarray,
100
+ key: str | None = None,
101
+ key_added: str = "eigenvalues_test",
102
+ flavor: Literal["asymptotic", "synthetic"] = "synthetic",
103
+ n_simulations: int = 10000,
104
+ progress_bar: bool = True,
105
+ null_distribution: np.ndarray | None = None,
106
+ ):
107
+ """
108
+ Performs Johnstone’s Spiked Covariance Test to identify if the embedding is random.
109
+
110
+ Parameters
111
+ ----------
112
+ adata : sc.AnnData | np.ndarray
113
+ AnnData object or numpy array containing the data.
114
+ key : str | None
115
+ Key to use for AnnData input.
116
+ key_added : str
117
+ Key to use for adding the results to the AnnData object.
118
+ flavor : Literal["asymptotic", "synthetic"]
119
+ Flavor of the test to use.
120
+ n_simulations : int
121
+ Number of simulations to use for the synthetic approach.
122
+ progress_bar : bool
123
+ Whether to show a progress bar.
124
+ null_distribution : np.ndarray | None
125
+ Null distribution to use for the synthetic approach.
126
+
127
+ Returns
128
+ -------
129
+ dict
130
+ Dictionary containing the test statistic and p-value.
131
+ """
132
+ start = logg.info(f"computing eigenvalues test via {flavor} approach")
133
+
134
+ if isinstance(adata, sc.AnnData):
135
+ if key is None:
136
+ raise ValueError("key must be specified for AnnData input")
137
+ X = adata.obsm[key]
138
+ elif isinstance(adata, np.ndarray):
139
+ X = adata
140
+ key_added = None
141
+ else:
142
+ raise ValueError("adata must be AnnData or numpy.ndarray")
143
+
144
+ if sc.settings.verbosity.value >= 2:
145
+ prefix = " "
146
+ progress_bar = True
147
+ else:
148
+ prefix = ""
149
+
150
+ stat = _calculate_statistics(X)
151
+ if flavor == "asymptotic":
152
+ p_value = 1.0 - _tracy_widom_cdf(stat)
153
+ elif flavor == "synthetic":
154
+ if null_distribution is None:
155
+ null_distribution = []
156
+ n, p = X.shape
157
+ cm = tqdm(
158
+ range(n_simulations),
159
+ desc=prefix + f"generating null distribution ({n_simulations} simulations)",
160
+ file=sys.stdout,
161
+ ) if progress_bar else nullcontext(range(n_simulations))
162
+ with cm:
163
+ for i in cm:
164
+ X_sim = np.random.normal(size=(n, p))
165
+ stat_sim = _calculate_statistics(X_sim)
166
+ null_distribution.append(stat_sim)
167
+ null_distribution = np.array(null_distribution)
168
+ p_value = np.mean(null_distribution >= stat)
169
+ if p_value == 0:
170
+ p_value = f"<{1 / n_simulations}"
171
+ else:
172
+ raise ValueError("flavor must be 'asymptotic' or 'synthetic'")
173
+
174
+ if key_added:
175
+ lines = [
176
+ "added",
177
+ f" .uns['{key_added}'] eigenvalues test statistics and approximate p-value",
178
+ ]
179
+ logg.info(" finished ({time_passed})", deep="\n".join([l for l in lines if l is not None]), time=start)
180
+ adata.uns[key_added] = {
181
+ "stat": stat,
182
+ "p_value": p_value,
183
+ "null_distribution": null_distribution,
184
+ }
185
+ else:
186
+ logg.info(" finished ({time_passed})", time=start)
187
+ return {"stat": stat, "p_value": p_value, "null_distribution": null_distribution}