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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279 | def call_reference_spots(config: dict, nbp_file: NotebookPage, nbp_basic: NotebookPage, nbp_ref_spots: NotebookPage,
hist_values: np.ndarray, hist_counts: np.ndarray,
transform: np.ndarray, overwrite_ref_spots: bool = False) -> Tuple[NotebookPage, NotebookPage]:
"""
This produces the bleed matrix and expected code for each gene as well as producing a gene assignment based on a
simple dot product for spots found on the reference round.
Returns the `call_spots` notebook page and adds the following variables to the `ref_spots` page:
`gene_no`, `score`, `score_diff`, `intensity`.
See `'call_spots'` and `'ref_spots'` sections of `notebook_comments.json` file
for description of the variables in each page.
Args:
config: Dictionary obtained from `'call_spots'` section of config file.
nbp_file: `file_names` notebook page
nbp_basic: `basic_info` notebook page
nbp_ref_spots: `ref_spots` notebook page containing all variables produced in `pipeline/reference_spots.py` i.e.
`local_yxz`, `isolated`, `tile`, `colors`.
`gene_no`, `score`, `score_diff`, `intensity` should all be `None` to add them here, unless
`overwrite_ref_spots == True`.
hist_values: `int [n_pixel_values]`.
All possible pixel values in saved tiff images i.e. `n_pixel_values` is approximately
`np.iinfo(np.uint16).max` because tiffs saved as `uint16` images.
This is saved in the extract notebook page i.e. `nb.extract.hist_values`.
hist_counts: `int [n_pixel_values x n_rounds x n_channels]`.
`hist_counts[i, r, c]` is the number of pixels across all tiles in round `r`, channel `c`
which had the value `hist_values[i]`.
This is saved in extract notebook page i.e. `nb.extract.hist_counts`.
transform: `float [n_tiles x n_rounds x n_channels x 4 x 3]`.
`transform[t, r, c]` is the affine transform to get from tile `t`, `ref_round`, `ref_channel` to
tile `t`, round `r`, channel `c`.
This is saved in the register notebook page i.e. `nb.register.transform`.
overwrite_ref_spots: If `True`, the variables:
* `gene_no`
* `score`
* `score_diff`
* `intensity`
in `nbp_ref_spots` will be overwritten if they exist. If this is `False`, they will only be overwritten
if they are all set to `None`, otherwise an error will occur.
Returns:
`NotebookPage[call_spots]` - Page contains bleed matrix and expected code for each gene.
`NotebookPage[ref_spots]` - Page contains gene assignments and info for spots found on reference round.
Parameters added are: intensity, score, gene_no, score_diff
"""
if overwrite_ref_spots:
warnings.warn("\noverwrite_ref_spots = True so will overwrite:\ngene_no, score, score_diff, intensity"
"\nin nbp_ref_spots.")
else:
# Raise error if data in nbp_ref_spots already exists that will be overwritted in this function.
error_message = ""
for var in ['gene_no', 'score', 'score_diff', 'intensity']:
if hasattr(nbp_ref_spots, var) and nbp_ref_spots.__getattribute__(var) is not None:
error_message += f"\nnbp_ref_spots.{var} is not None but this function will overwrite {var}." \
f"\nRun with overwrite_ref_spots = True to get past this error."
if len(error_message) > 0:
raise ValueError(error_message)
nbp_ref_spots.finalized = False # So we can add and delete ref_spots page variables
# delete all variables in ref_spots set to None so can add them later.
for var in ['gene_no', 'score', 'score_diff', 'intensity']:
if hasattr(nbp_ref_spots, var):
nbp_ref_spots.__delattr__(var)
nbp = NotebookPage("call_spots")
# get color norm factor
rc_ind = np.ix_(nbp_basic.use_rounds, nbp_basic.use_channels)
hist_counts_use = np.moveaxis(np.moveaxis(hist_counts, 0, -1)[rc_ind], -1, 0)
color_norm_factor = np.ones((nbp_basic.n_rounds, nbp_basic.n_channels)) * np.nan
color_norm_factor[rc_ind] = color_normalisation(hist_values, hist_counts_use, config['color_norm_intensities'],
config['color_norm_probs'], config['bleed_matrix_method'])
# get initial bleed matrix
initial_raw_bleed_matrix = np.ones((nbp_basic.n_rounds, nbp_basic.n_channels, nbp_basic.n_dyes)) * np.nan
rcd_ind = np.ix_(nbp_basic.use_rounds, nbp_basic.use_channels, nbp_basic.use_dyes)
if nbp_basic.dye_names is not None:
# if specify dyes, will initialize bleed matrix using prior data
dye_names_use = np.array(nbp_basic.dye_names)[nbp_basic.use_dyes]
camera_use = np.array(nbp_basic.channel_camera)[nbp_basic.use_channels]
laser_use = np.array(nbp_basic.channel_laser)[nbp_basic.use_channels]
initial_raw_bleed_matrix[rcd_ind] = get_dye_channel_intensity_guess(nbp_file.dye_camera_laser,
dye_names_use, camera_use,
laser_use).transpose()
initial_bleed_matrix = initial_raw_bleed_matrix / np.expand_dims(color_norm_factor, 2)
else:
if nbp_basic.n_dyes != nbp_basic.n_channels:
raise ValueError(f"'dye_names' were not specified so expect each dye to correspond to a different channel."
f"\nBut n_channels={nbp_basic.n_channels} and n_dyes={nbp_basic.n_dyes}")
if nbp_basic.use_channels != nbp_basic.use_dyes:
raise ValueError(f"'dye_names' were not specified so expect each dye to correspond to a different channel."
f"\nBleed matrix computation requires use_channels and use_dyes to be the same to work."
f"\nBut use_channels={nbp_basic.use_channels} and use_dyes={nbp_basic.use_dyes}")
initial_bleed_matrix = initial_raw_bleed_matrix.copy()
initial_bleed_matrix[rcd_ind] = np.tile(np.expand_dims(np.eye(nbp_basic.n_channels), 0),
(nbp_basic.n_rounds, 1, 1))[rcd_ind]
# Get norm_shift and intensity_thresh from middle tile/ z-plane average intensity
# This is because these variables are all a small fraction of a spot_color L2 norm in one round.
# Hence, use average pixel as example of low intensity spot.
# get central tile
nbp.norm_shift_tile = scale.central_tile(nbp_basic.tilepos_yx, nbp_basic.use_tiles)
if nbp_basic.is_3d:
nbp.norm_shift_z = int(np.floor(nbp_basic.nz / 2)) # central z-plane to get info from.
else:
nbp.norm_shift_z = 0
pixel_colors = get_spot_colors(all_pixel_yxz(nbp_basic.tile_sz, nbp_basic.tile_sz, nbp.norm_shift_z),
nbp.norm_shift_tile, transform, nbp_file, nbp_basic, return_in_bounds=True)[0]
pixel_intensity = get_spot_intensity(np.abs(pixel_colors) / color_norm_factor[rc_ind])
nbp.abs_intensity_percentile = np.percentile(pixel_intensity, np.arange(1, 101))
if config['background_weight_shift'] is None:
# Set to median absolute pixel intensity
config['background_weight_shift'] = float(round_any(nbp.abs_intensity_percentile[50],
config['norm_shift_precision'], 'ceil'))
median_round_l2_norm = np.median(np.linalg.norm(pixel_colors / color_norm_factor[rc_ind], axis=2))
if config['dp_norm_shift'] is None:
config['dp_norm_shift'] = float(round_any(median_round_l2_norm, config['norm_shift_precision']))
# intensity thresh is just a very low threshold, would basically be the same if set to 0
# but found it to be slightly better on ground truth
pixel_intensity = get_spot_intensity(pixel_colors / color_norm_factor[rc_ind])
if config['gene_efficiency_intensity_thresh'] is None:
config['gene_efficiency_intensity_thresh'] = \
float(round_any(np.percentile(pixel_intensity, config['gene_efficiency_intensity_thresh_percentile']),
config['gene_efficiency_intensity_thresh_precision']))
nbp.dp_norm_shift = float(np.clip(config['dp_norm_shift'], config['norm_shift_min'], config['norm_shift_max']))
nbp.background_weight_shift = float(np.clip(config['background_weight_shift'],
config['norm_shift_min'], config['norm_shift_max']))
nbp.gene_efficiency_intensity_thresh = \
float(np.clip(config['gene_efficiency_intensity_thresh'],
config['gene_efficiency_intensity_thresh_min'],
config['gene_efficiency_intensity_thresh_max']))
# get bleed matrix
spot_colors_use = np.moveaxis(np.moveaxis(nbp_ref_spots.colors, 0, -1)[rc_ind], -1, 0) / color_norm_factor[rc_ind]
nbp_ref_spots.intensity = np.asarray(get_spot_intensity(spot_colors_use).astype(np.float32))
# Remove background first
background_coef = np.ones((spot_colors_use.shape[0], nbp_basic.n_channels)) * np.nan
background_codes = np.ones((nbp_basic.n_channels, nbp_basic.n_rounds, nbp_basic.n_channels)) * np.nan
crc_ind = np.ix_(nbp_basic.use_channels, nbp_basic.use_rounds, nbp_basic.use_channels)
spot_colors_use, background_coef[:, nbp_basic.use_channels], background_codes[crc_ind] = \
fit_background(spot_colors_use, nbp.background_weight_shift)
spot_colors_use = np.asarray(spot_colors_use) # in case using jax
bleed_matrix = initial_raw_bleed_matrix.copy()
bleed_matrix[rcd_ind] = get_bleed_matrix(spot_colors_use[nbp_ref_spots.isolated], initial_bleed_matrix[rcd_ind],
config['bleed_matrix_method'], config['bleed_matrix_score_thresh'],
config['bleed_matrix_min_cluster_size'], config['bleed_matrix_n_iter'],
config['bleed_matrix_anneal'])
# get gene codes
gene_names, gene_codes = np.genfromtxt(nbp_file.code_book, dtype=(str, str)).transpose()
gene_codes = np.array([[int(i) for i in gene_codes[j]] for j in range(len(gene_codes))])
# bled_codes[g,r,c] returned below is nan where r/c/gene_codes[g,r] outside use_rounds/channels/dyes
bled_codes = get_bled_codes(gene_codes, bleed_matrix)
# get bled_codes_use with no nan values and L2 norm=1
bled_codes_use = np.moveaxis(np.moveaxis(bled_codes, 0, -1)[rc_ind], -1, 0)
bled_codes_use[np.isnan(bled_codes_use)] = 0 # set all round vectors where dye is not in use_dyes to 0.
# Give all bled codes an L2 norm of 1 across use_rounds and use_channels
norm_factor = np.expand_dims(np.linalg.norm(bled_codes_use, axis=(1, 2)), (1, 2))
norm_factor[norm_factor == 0] = 1 # For genes with no dye in use_dye, this avoids blow up on next line
bled_codes_use = bled_codes_use / norm_factor
# bled_codes[g,r,c] so nan when r/c outside use_rounds/channels and 0 when gene_codes[g,r] outside use_dyes
n_genes = bled_codes_use.shape[0]
bled_codes = np.ones((nbp_basic.n_rounds, nbp_basic.n_channels, n_genes)) * np.nan
bled_codes[rc_ind] = np.moveaxis(bled_codes_use, 0, -1)
bled_codes = np.moveaxis(bled_codes, -1, 0)
nbp.gene_names = gene_names
nbp.gene_codes = gene_codes
nbp.color_norm_factor = color_norm_factor
nbp.initial_raw_bleed_matrix = initial_raw_bleed_matrix
nbp.initial_bleed_matrix = initial_bleed_matrix
nbp.bleed_matrix = bleed_matrix
nbp.bled_codes = bled_codes
nbp.background_codes = background_codes
# shift in config file is just for one round.
n_spots, n_rounds_use, n_channels_use = spot_colors_use.shape
dp_norm_shift = nbp.dp_norm_shift * np.sqrt(n_rounds_use)
# Down-weight round/channels with high background when compute dot product
alpha = config['alpha']
beta = config['beta']
background_codes = background_codes[crc_ind].reshape(n_channels_use, -1)
background_var = background_coef[:, nbp_basic.use_channels]**2 @ background_codes**2 * alpha + beta ** 2
# find spot assignments to genes and gene efficiency
n_iter = config['gene_efficiency_n_iter'] + 1
pass_intensity_thresh = nbp_ref_spots.intensity > nbp.gene_efficiency_intensity_thresh
use_ge_last = np.zeros(n_spots).astype(bool)
bled_codes_ge_use = bled_codes_use.copy()
for i in range(n_iter):
scores = np.asarray(dot_product_score(spot_colors_use.reshape(n_spots, -1),
bled_codes_ge_use.reshape(n_genes, -1), dp_norm_shift, 1/background_var))
spot_gene_no = np.argmax(scores, 1)
spot_score = scores[np.arange(np.shape(scores)[0]), spot_gene_no]
pass_score_thresh = spot_score > config['gene_efficiency_score_thresh']
sort_gene_inds = np.argsort(scores, axis=1)
gene_no_second_best = sort_gene_inds[:, -2]
score_second_best = scores[np.arange(np.shape(scores)[0]), gene_no_second_best]
spot_score_diff = spot_score - score_second_best
pass_score_diff_thresh = spot_score_diff > config['gene_efficiency_score_diff_thresh']
# only use isolated spots which pass strict thresholding to compute gene_efficiencies
use_ge = np.array([nbp_ref_spots.isolated, pass_intensity_thresh, pass_score_thresh,
pass_score_diff_thresh]).all(axis=0)
# nan_to_num line below converts nan in bleed_matrix to 0.
# This basically just says that for dyes not in use_dyes, we expect intensity to be 0.
gene_efficiency_use = get_gene_efficiency(spot_colors_use[use_ge], spot_gene_no[use_ge],
gene_codes[:, nbp_basic.use_rounds],
np.nan_to_num(bleed_matrix[rc_ind]),
config['gene_efficiency_min_spots'],
config['gene_efficiency_max'],
config['gene_efficiency_min'],
config['gene_efficiency_min_factor'])
# get new bled codes using gene efficiency with L2 norm = 1.
multiplier_ge = np.tile(np.expand_dims(gene_efficiency_use, 2), [1, 1, n_channels_use])
bled_codes_ge_use = bled_codes_use * multiplier_ge
norm_factor = np.expand_dims(np.linalg.norm(bled_codes_ge_use, axis=(1, 2)), (1, 2))
norm_factor[norm_factor == 0] = 1 # For genes with no dye in use_dye, this avoids blow up on next line
bled_codes_ge_use = bled_codes_ge_use / norm_factor
if np.sum(use_ge != use_ge_last) < 3:
# if less than 3 spots different in spots used for ge computation, end.
break
use_ge_last = use_ge.copy()
if config['gene_efficiency_n_iter'] > 0:
# Compute score with final gene efficiency
scores = np.asarray(dot_product_score(spot_colors_use.reshape(n_spots, -1),
bled_codes_ge_use.reshape(n_genes, -1), dp_norm_shift, 1/background_var))
spot_gene_no = np.argmax(scores, 1)
spot_score = scores[np.arange(np.shape(scores)[0]), spot_gene_no]
else:
bled_codes_ge_use = bled_codes_use.copy()
# save score using the latest gene efficiency and diff to second best gene
nbp_ref_spots.score = spot_score.astype(np.float32)
nbp_ref_spots.gene_no = spot_gene_no.astype(np.int16)
sort_gene_inds = np.argsort(scores, axis=1)
gene_no_second_best = sort_gene_inds[:, -2]
score_second_best = scores[np.arange(np.shape(scores)[0]), gene_no_second_best]
nbp_ref_spots.score_diff = (nbp_ref_spots.score - score_second_best).astype(np.float16)
# save gene_efficiency[g,r] with nan when r outside use_rounds and 1 when gene_codes[g,r] outside use_dyes.
gene_efficiency = np.ones((n_genes, nbp_basic.n_rounds)) * np.nan
gene_efficiency[:, nbp_basic.use_rounds] = gene_efficiency_use
nbp.gene_efficiency = gene_efficiency
# bled_codes_ge[g,r,c] so nan when r/c outside use_rounds/channels and 0 when gene_codes[g,r] outside use_dyes
bled_codes_ge = np.ones((nbp_basic.n_rounds, nbp_basic.n_channels, n_genes)) * np.nan
bled_codes_ge[rc_ind] = np.moveaxis(bled_codes_ge_use, 0, -1)
bled_codes_ge = np.moveaxis(bled_codes_ge, -1, 0)
nbp.bled_codes_ge = bled_codes_ge
ge_fail_genes = np.where(np.min(gene_efficiency_use,axis=1) == 1)[0]
n_fail_ge = len(ge_fail_genes)
if n_fail_ge > 0:
fail_genes_str = [str(ge_fail_genes[i]) + ': ' + gene_names[ge_fail_genes][i] for i in range(n_fail_ge)]
fail_genes_str = '\n'.join(fail_genes_str)
warnings.warn(f"\nGene Efficiency could not be calculated for {n_fail_ge}/{n_genes} "
f"genes:\n{fail_genes_str}")
nbp_ref_spots.finalized = True
return nbp, nbp_ref_spots
|