File size: 14,663 Bytes
87da3f2 9a4e257 87da3f2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 | import shap
import pandas as pd
import os
from pathlib import Path
import seaborn as sns
import matplotlib.pyplot as plt
import torch
import numpy as np
import pickle
from sklearn.metrics import mean_absolute_error
from scipy.stats import pearsonr
from models.tabular.widedeep.ft_transformer import WDFTTransformerModel
import gradio as gr
from scipy import stats
import warnings
warnings.filterwarnings("ignore",
".*will save all targets and predictions in the buffer. For large datasets, this may lead to large memory footprint.*")
warnings.filterwarnings("ignore", ".*is non-interactive, and thus cannot be shown*")
root_dir = Path(os.getcwd())
fn_model = f"{root_dir}/data/model.ckpt"
model = WDFTTransformerModel.load_from_checkpoint(checkpoint_path=fn_model)
model.eval()
model.freeze()
feats = [
'CXCL9',
'CCL22',
'IL6',
'PDGFB',
'CD40LG',
'IL27',
'VEGFA',
'CSF1',
'PDGFA',
'CXCL10'
]
fn_shap = f"{root_dir}/data/shap.pickle"
out_dir = f"{root_dir}/out"
if not os.path.exists(out_dir):
os.makedirs(out_dir)
def predict_func(x):
batch = {
'all': torch.from_numpy(np.float32(x)),
'continuous': torch.from_numpy(np.float32(x)),
'categorical': torch.from_numpy(np.int32(x[:, []])),
}
return model(batch).cpu().detach().numpy()
with open(fn_shap, 'rb') as handle:
shap_dict = pickle.load(handle)
values_train = shap_dict['values_train']
shap_values_train = shap_dict['shap_values_train']
explainer = shap_dict['explainer']
def predict(input):
if input.endswith('xlsx'):
df = pd.read_excel(input, index_col=0)
elif input.endswith('csv'):
df = pd.read_csv(input, index_col=0)
else:
raise gr.Error(f"Unknown file type!")
if "Age" not in df.columns:
raise gr.Error("No 'Age' column in the input file!")
missed_features = [feature for feature in feats if feature not in df.columns]
if len(missed_features) > 0:
raise gr.Error(f"No {', '.join(missed_features)} column(s) in the input file!")
try:
df = df.loc[:, feats + ['Age']]
except ValueError:
raise gr.Error(f"Non-numeric value in 'Age' column!")
df = df.astype({'Age': 'float'})
for feat in feats:
try:
df = df.astype({feat: 'float'})
except ValueError:
raise gr.Error(f"Non-numeric value in '{feat}' column!")
df['SImAge'] = model(torch.from_numpy(df.loc[:, feats].values)).cpu().detach().numpy().ravel()
df['SImAge acceleration'] = df['SImAge'] - df['Age']
df.to_excel(f'{root_dir}/out/df.xlsx')
df_res = df[['SImAge']]
df_res.to_excel(f'{root_dir}/out/result.xlsx')
if len(df) > 1:
mae = mean_absolute_error(df['Age'].values, df['SImAge'].values)
rho = pearsonr(df['Age'].values, df['SImAge'].values).statistic
plt.close('all')
sns.set_theme(style='whitegrid')
fig, ax = plt.subplots(figsize=(4, 4))
scatter = sns.scatterplot(
data=df,
x="Age",
y="SImAge",
linewidth=0.1,
alpha=0.75,
edgecolor="k",
s=40,
color='blue',
ax=ax
)
bisect = sns.lineplot(
x=[0, 120],
y=[0, 120],
linestyle='--',
color='black',
linewidth=1.0,
ax=ax
)
ax.set_xlim(0, 120)
ax.set_ylim(0, 120)
plt.savefig(f'{root_dir}/out/scatter.svg', bbox_inches='tight')
plt.close('all')
if len(df) > 1:
sns.set_theme(style='whitegrid')
fig, ax = plt.subplots(figsize=(2, 4))
sns.violinplot(
data=df,
y='SImAge acceleration',
density_norm='width',
color='blue',
saturation=0.75,
)
plt.savefig(f'{root_dir}/out/violin.svg', bbox_inches='tight')
plt.close('all')
shap.summary_plot(
shap_values=shap_values_train.values,
features=values_train.values,
feature_names=feats,
max_display=len(feats),
plot_type="violin",
)
plt.savefig(f'{root_dir}/out/shap_beeswarm.svg', bbox_inches='tight')
plt.close('all')
if len(df) > 1:
return_metrics = gr.update(value=f'MAE: {round(mae, 3)}\nPearson Rho: {round(rho, 3)}', visible=True)
return_gallery = gr.update(value=[(f'{root_dir}/out/scatter.svg', 'Scatter'),
(f'{root_dir}/out/violin.svg', 'Violin'),
(f'{root_dir}/out/shap_beeswarm.svg', 'SHAP Beeswarm')], visible=True)
else:
return_metrics = gr.update(value=f'Only one sample.\nNo metrics can be calculated.', visible=True)
return_gallery = gr.update(value=[(f'{root_dir}/out/scatter.svg', 'Scatter'),
(f'{root_dir}/out/shap_beeswarm.svg', 'SHAP Beeswarm')], visible=True)
return [return_metrics,
gr.update(value=f'{root_dir}/out/result.xlsx', visible=True),
return_gallery,
gr.update(visible=True), gr.update(visible=True),
gr.update(choices=list(df.index.values), value=list(df.index.values)[0], interactive=True, visible=True),
gr.update(visible=True), gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)]
def explain(input):
df = pd.read_excel(f'{root_dir}/out/df.xlsx', index_col=0)
trgt_id = input
shap_values_trgt = explainer.shap_values(df.loc[trgt_id, feats].values)
base_value = explainer.expected_value[0]
age = df.loc[trgt_id, ['Age']].values[0]
simage = df.loc[trgt_id, ['SImAge']].values[0]
order = np.argsort(-np.abs(shap_values_trgt))
locally_ordered_feats = [feats[i] for i in order]
plt.close('all')
shap.plots.waterfall(
shap.Explanation(
values=shap_values_trgt,
base_values=base_value,
data=df.loc[trgt_id, feats].values,
feature_names=feats
),
max_display=len(feats),
show=True,
)
plt.savefig(f'{root_dir}/out/waterfall_{trgt_id}.svg', bbox_inches='tight')
plt.close('all')
if len(df) > 1:
age_window = 5
trgt_age = df.at[trgt_id, 'Age']
trgt_simage = df.at[trgt_id, 'SImAge']
trgt_simage_acc = df.at[trgt_id, 'SImAge acceleration']
ids_near = df.index[(df['Age'] >= trgt_age - age_window) & (df['Age'] < trgt_age + age_window)]
trgt_simage_acc_prctl = stats.percentileofscore(df.loc[ids_near, 'SImAge acceleration'], trgt_simage_acc)
sns.set(style='whitegrid', font_scale=1.5)
fig, ax = plt.subplots(figsize=(10, 6))
kdeplot = sns.kdeplot(
data=df.loc[ids_near, :],
x='SImAge acceleration',
color='gray',
linewidth=4,
cut=0,
ax=ax
)
kdeline = ax.lines[0]
xs = kdeline.get_xdata()
ys = kdeline.get_ydata()
ax.fill_between(xs, 0, ys, where=(xs <= trgt_simage_acc), interpolate=True, facecolor='dodgerblue', alpha=0.7)
ax.fill_between(xs, 0, ys, where=(xs >= trgt_simage_acc), interpolate=True, facecolor='crimson', alpha=0.7)
ax.vlines(trgt_simage_acc, 0, np.interp(trgt_simage_acc, xs, ys), color='black', linewidth=6)
ax.text(np.mean([min(xs), trgt_simage_acc]), 0.1 * max(ys), f"{trgt_simage_acc_prctl:0.1f}%",
fontstyle="oblique",
color="black", ha="center", va="center")
ax.text(np.mean([max(xs), trgt_simage_acc]), 0.1 * max(ys), f"{100 - trgt_simage_acc_prctl:0.1f}%",
fontstyle="oblique", color="black", ha="center", va="center")
fig.savefig(f"{root_dir}/out/kde_aa_{trgt_id}.svg", bbox_inches='tight')
plt.close(fig)
sns.set(style='whitegrid', font_scale=0.7)
n_rows = 2
n_cols = 5
fig_height = 4
fig_width = 10
fig, axs = plt.subplots(n_rows, n_cols, figsize=(fig_width, fig_height), gridspec_kw={}, sharey=False,
sharex=False)
for feat_id, feat in enumerate(feats):
row_id, col_id = divmod(feat_id, n_cols)
kdeplot = sns.kdeplot(
data=df.loc[ids_near, :],
x=feat,
color='gray',
linewidth=1,
cut=0,
ax=axs[row_id, col_id]
)
kdeline = axs[row_id, col_id].lines[0]
xs = kdeline.get_xdata()
ys = kdeline.get_ydata()
trgt_val = df.at[trgt_id, feat]
trgt_prctl = stats.percentileofscore(df.loc[ids_near, feat], trgt_val)
axs[row_id, col_id].fill_between(xs, 0, ys, where=(xs <= trgt_val), interpolate=True,
facecolor='dodgerblue',
alpha=0.7)
axs[row_id, col_id].fill_between(xs, 0, ys, where=(xs >= trgt_val), interpolate=True, facecolor='crimson',
alpha=0.7)
axs[row_id, col_id].vlines(trgt_val, 0, np.interp(trgt_val, xs, ys), color='black', linewidth=1.5)
axs[row_id, col_id].text(np.mean([min(xs), trgt_val]), 0.1 * max(ys), f"{trgt_prctl:0.1f}%",
fontstyle="oblique",
color="black", ha="center", va="center")
axs[row_id, col_id].text(np.mean([max(xs), trgt_val]), 0.1 * max(ys), f"{100 - trgt_prctl:0.1f}%",
fontstyle="oblique",
color="black", ha="center", va="center")
axs[row_id, col_id].ticklabel_format(style='scientific', scilimits=(-1, 1), axis='y', useOffset=True)
fig.tight_layout()
fig.savefig(f"{root_dir}/out/kde_feats_{trgt_id}.svg", bbox_inches='tight')
plt.close(fig)
if len(df) > 1:
return_gallery = [(f'{root_dir}/out/waterfall_{trgt_id}.svg', 'Waterfall'),
(f'{root_dir}/out/kde_aa_{trgt_id}.svg', 'Age Acceleration KDE'),
(f'{root_dir}/out/kde_feats_{trgt_id}.svg', 'Features KDE')]
else:
return_gallery = [(f'{root_dir}/out/waterfall_{trgt_id}.svg', 'Waterfall')]
return [f'Real age: {round(age, 3)}\nSImAge: {round(simage, 3)}',
f'{locally_ordered_feats[0]}\n{locally_ordered_feats[1]}\n{locally_ordered_feats[2]}',
return_gallery]
def clear():
return (gr.update(interactive=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False),
gr.update(value=None, visible=False), gr.update(value=None, visible=False),
gr.update(value=None, visible=False))
def check_size(input):
curr_file_size = os.path.getsize(input)
if curr_file_size > 1024 * 1024:
raise gr.Error(f"File exceeds 1 MB limit!")
else:
return gr.update(interactive=True)
css = """
h2 {
text-align: center;
display:block;
}
"""
with gr.Blocks(css=css, theme=gr.themes.Soft(), title='SImAge') as app:
gr.Markdown(
"""
<h2>Calculate your immunological age using SImAge model</h2>
"""
)
with gr.Row():
with gr.Column():
gr.Markdown(
"""
### Submit immunology data
The file should contain chronological age ("Age" column) and immunology data for the following 10 cytokines:
CXCL9, CCL22, IL6, PDGFB, CD40LG, IL27, VEGFA, CSF1, PDGFA, CXCL10
"""
)
input_file = gr.File(label='Input file', file_count='single', file_types=['.xlsx', 'csv'])
submit_button = gr.Button("Submit data", variant="primary", interactive=False)
with gr.Column():
with gr.Row():
output_text = gr.Text(label='Main metrics', visible=False)
output_file = gr.File(label='Result file', file_types=['.xlsx'], interactive=False, visible=False)
with gr.Row():
gallery = gr.Gallery(label='Figures Gallery', object_fit='cover', columns=2, rows=2, visible=False)
title_shap = gr.Markdown(
"""
<h2>Local explainability</h2>
"""
, visible=False)
with gr.Row():
with gr.Column():
text_shap = gr.Markdown(
"""
Select a record to get an explanation of the SImAge prediction:
"""
, visible=False)
input_shap = gr.Dropdown(label='Choose a sample', visible=False)
shap_button = gr.Button("Get explanation", variant="primary", visible=False)
with gr.Column():
with gr.Row():
with gr.Column(scale=1):
shap_local = gr.Text(label='Sample info', visible=False)
shap_cyto = gr.Text(label='Most important cytokines', visible=False)
with gr.Column(scale=3):
shap_gallery = gr.Gallery(label='Local Explainability Gallery', object_fit='cover', columns=2,
rows=2, visible=False)
submit_button.click(fn=predict,
inputs=[input_file],
outputs=[output_text, output_file, gallery, title_shap, text_shap, input_shap, shap_button,
shap_local,
shap_cyto, shap_gallery]
)
shap_button.click(fn=explain,
inputs=[input_shap],
outputs=[shap_local, shap_cyto, shap_gallery]
)
input_file.clear(fn=clear,
inputs=[],
outputs=[submit_button, output_text, output_file, gallery,
title_shap, text_shap, input_shap, shap_button, shap_local, shap_cyto, shap_gallery])
input_file.upload(fn=check_size,
inputs=[input_file],
outputs=[submit_button])
gr.Markdown(
"""
Reference:
Kalyakulina, A., Yusipov, I., Kondakova, E., Bacalini, M. G., Franceschi, C., Vedunova, M., & Ivanchenko, M. (2023). [Small immunological clocks identified by deep learning and gradient boosting](https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2023.1177611/full). Frontiers in Immunology, 14, 1177611.
"""
)
app.launch()
|