Gut–brain axis hierarchical analysis in autism spectrum disorder (ASD)#
This notebook applies CauTrigger to model hierarchical regulation along the gut–brain axis, from gut microbiota → host proteome → ASD-related neurodevelopmental states.
Steps include:
Identifying causal proteins associated with ASD from the host proteome layer.
Inferring upstream microbiota contributing to host variation and ASD states.
Performing in silico activation of key taxa (e.g., Faecalibacterium) to simulate microbiota-driven modulation.
Data source：Integrative multi-omics analysis of autism spectrum disorder reveals unique microbial macromolecules interactions.
Import and environment setup#
import os
import math
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from itertools import combinations
from sklearn.feature_selection import mutual_info_classif
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import normalize, minmax_scale
from scipy.stats import pearsonr, spearmanr, f_oneway, norm
import warnings
warnings.filterwarnings("ignore")
plt.rcParams['pdf.fonttype'] = 42
plt.rcParams['font.sans-serif'] = ['Arial']
plt.rcParams['font.family'] = 'sans-serif'
import logging
logging.getLogger('matplotlib.font_manager').disabled = True
import scanpy as sc
import torch
import torch.nn as nn
torch.cuda.set_device(0)
import re
from sklearn.preprocessing import FunctionTransformer
from cautrigger.utils import set_seed, pert_plot_stream
from cautrigger.model import CauTrigger1L, CauTrigger2L
Load and preprocess data#
data_dir = '../data/'
output_dir = '../output/'
os.makedirs(output_dir, exist_ok=True)
microb_df = pd.read_csv(os.path.join(data_dir, '16sRNA.csv'), index_col=0).fillna(0)
microb_df = microb_df.iloc[np.where(np.sum(microb_df.T != 0) >= 6)[0], :]
microb = minmax_scale(microb_df)
microb_df.iloc[:] = microb
prot = pd.read_csv(os.path.join(data_dir, 'host_Proteomics.csv'), index_col=0)
def reorder_column_name(col):
match = re.match(r"(\d+)([A-Z])", col)
if match:
return match.group(2) + match.group(1)
else:
return col
prot.columns = [reorder_column_name(col) for col in prot.columns]
def sort_key(col):
match = re.match(r"([A-Z])(\d+)", col)
if match:
return (match.group(1), int(match.group(2)))
return (col, 0)
sorted_columns = sorted(prot.columns, key=sort_key)
prot = prot[sorted_columns]
labels = prot.loc['Label']
prot_df = prot.iloc[1:]
prot = minmax_scale(prot_df)
prot_df.iloc[:] = prot
Train CauTrigger (Microbiota → Proteome → ASD)#
set_seed(42)
adata_layer1 = sc.AnnData(X=np.array(prot, dtype=np.float32).T)
adata_layer1.obs_names = prot_df.columns
adata_layer1.var_names = prot_df.index
adata_layer1.obs['labels_ori'] = labels.values
adata_layer1.obs['labels'] = labels.apply(lambda tg: 1 if tg == labels[0] else 0)
model_1L = CauTrigger1L(
adata_layer1,
n_latent=10,
n_hidden=128,
n_layers_encoder=0,
n_layers_decoder=0,
n_layers_dpd=0,
dropout_rate_encoder=0.0,
dropout_rate_decoder=0.0,
dropout_rate_dpd=0.0,
use_batch_norm='none',
use_batch_norm_dpd=True,
decoder_linear=True,
dpd_linear=False,
init_weight=None,
init_thresh=0.0,
attention=False,
att_mean=False,
)
model_1L.train(max_epochs=300, stage_training=True)
df_layer1, _ = model_1L.get_up_feature_weights(method='SHAP', sort_by_weight=False)
topk_indices = df_layer1["weight"].values.argsort()[-100:]
X_down = adata_layer1[:, topk_indices]
adata_layer2 = sc.AnnData(
X=microb.T,
obsm={"X_down": X_down.X.toarray()}
)
adata_layer2.obs_names = microb_df.columns
adata_layer2.var_names = microb_df.index
adata_layer2.obs['labels_ori'] = labels.values
adata_layer2.obs['labels'] = labels.apply(lambda tg: 1 if tg == labels[0] else 0)
model_2L = CauTrigger2L(
adata_layer2,
n_latent=10,
n_hidden=128,
n_layers_encoder=0,
n_layers_decoder=0,
n_layers_dpd=0,
dropout_rate_encoder=0.0,
dropout_rate_decoder=0.0,
dropout_rate_dpd=0.0,
use_batch_norm='none',
use_batch_norm_dpd=True,
decoder_linear=True,
dpd_linear=True,
init_weight=None,
init_thresh=0.0,
attention=False,
att_mean=False,
)
model_2L.train(max_epochs=300, stage_training=True)
df_layer2, _ = model_2L.get_up_feature_weights(method='SHAP', sort_by_weight=True)
Information flow and downstream analyses#
model_1L.compute_information_flow(plot_info_flow=True, save_fig=True, save_dir=output_dir)
( 0 1 2 3 4 5 6 \
A1 0.524907 0.451544 0.002385 0.004177 0.002489 0.001596 0.004022
A2 0.597220 0.364063 0.005251 0.006858 0.002189 0.002709 0.005120
A3 0.457608 0.516787 0.002057 0.003200 0.001806 0.002857 0.002893
A4 0.509703 0.461390 0.003265 0.004623 0.002165 0.002186 0.004340
A5 0.598944 0.381025 0.003355 0.003202 0.001371 0.002343 0.001705
A6 0.739513 0.251651 0.001365 0.002118 0.000341 0.000453 0.000590
A7 0.598032 0.388858 0.003271 0.002092 0.000582 0.000569 0.000979
A8 0.394622 0.575781 0.002056 0.003695 0.003354 0.003242 0.003407
A9 0.616491 0.380930 0.000464 0.000402 0.000180 0.000092 0.000291
A10 0.674995 0.297843 0.003504 0.006259 0.001081 0.000890 0.001446
A11 0.662705 0.327885 0.002326 0.002798 0.000401 0.000703 0.000456
A12 0.479497 0.513900 0.000812 0.001296 0.000524 0.000338 0.000470
A13 0.506226 0.474587 0.001150 0.003126 0.002709 0.001463 0.002801
A14 0.514751 0.449148 0.003233 0.008091 0.003575 0.002944 0.003039
A15 0.545012 0.420327 0.002756 0.008703 0.001983 0.002700 0.002692
A16 0.515123 0.463639 0.001529 0.003154 0.002396 0.002173 0.002414
A17 0.517114 0.453305 0.003447 0.003490 0.003144 0.002575 0.002526
A18 0.731172 0.236862 0.002316 0.006029 0.003218 0.005663 0.003136
A19 0.608099 0.363016 0.004199 0.005422 0.002667 0.001966 0.002288
A20 0.564350 0.392190 0.005515 0.006580 0.004226 0.002846 0.006098
A21 0.581593 0.382785 0.004876 0.008387 0.003283 0.003287 0.006357
A22 0.451459 0.520891 0.003442 0.003021 0.002460 0.002613 0.003111
A23 0.478778 0.494034 0.004856 0.003241 0.001634 0.002632 0.002445
A24 0.467127 0.500384 0.003640 0.007487 0.004517 0.003520 0.002614
A25 0.828035 0.166798 0.000668 0.001723 0.000271 0.000205 0.000347
A26 0.586770 0.378625 0.004891 0.006191 0.006349 0.003232 0.003072
A27 0.535537 0.427484 0.004602 0.004798 0.005901 0.004095 0.004803
A28 0.645411 0.297264 0.007508 0.006265 0.008294 0.005999 0.006878
A29 0.491445 0.480606 0.002988 0.005452 0.003734 0.003365 0.003354
A30 0.501783 0.472566 0.002253 0.003213 0.001808 0.002841 0.004887
C1 0.469132 0.509451 0.002599 0.004121 0.002822 0.002628 0.002282
C2 0.640206 0.359567 0.000015 0.000028 0.000022 0.000019 0.000037
C3 0.497838 0.497447 0.000436 0.000782 0.000525 0.000458 0.000714
C4 0.559726 0.438930 0.000111 0.000177 0.000193 0.000165 0.000214
C5 0.555440 0.439174 0.000359 0.001122 0.000675 0.000636 0.000531
C6 0.508630 0.490838 0.000097 0.000063 0.000078 0.000043 0.000076
C7 0.591296 0.400860 0.000537 0.001528 0.000905 0.000519 0.001621
C8 0.518065 0.455951 0.003173 0.004190 0.002119 0.001928 0.003501
C9 0.311615 0.665736 0.002836 0.004116 0.002229 0.002517 0.003873
C10 0.470567 0.506557 0.001379 0.004261 0.002567 0.001787 0.001789
C11 0.598464 0.401480 0.000003 0.000008 0.000009 0.000008 0.000009
C12 0.542859 0.456904 0.000014 0.000023 0.000037 0.000045 0.000026
C13 0.496678 0.489451 0.001788 0.002987 0.002657 0.000702 0.001487
C14 0.518680 0.474970 0.000576 0.001245 0.001086 0.000672 0.000699
C15 0.511998 0.478814 0.000835 0.001491 0.000871 0.001291 0.001581
C16 0.458549 0.533879 0.000823 0.000607 0.001438 0.000950 0.000722
C17 0.471724 0.527348 0.000059 0.000110 0.000205 0.000078 0.000080
C18 0.533720 0.458726 0.000843 0.001375 0.000850 0.001235 0.000995
C19 0.538224 0.451261 0.000675 0.001354 0.001544 0.001259 0.001573
C20 0.499987 0.499520 0.000021 0.000068 0.000099 0.000041 0.000057
C21 0.499359 0.495956 0.000675 0.000650 0.000746 0.000358 0.000451
C22 0.532167 0.463187 0.000387 0.000754 0.000514 0.000395 0.000836
C23 0.489523 0.487323 0.002786 0.003353 0.002504 0.003293 0.002382
C24 0.457938 0.540194 0.000078 0.000239 0.000234 0.000166 0.000310
C25 0.598887 0.382977 0.002054 0.002898 0.003233 0.001840 0.001724
C26 0.537830 0.445050 0.001861 0.002007 0.003601 0.001544 0.001989
C27 0.463439 0.524999 0.001121 0.002082 0.001393 0.001650 0.001125
C28 0.493460 0.490080 0.000852 0.003186 0.001768 0.001943 0.001585
C29 0.464190 0.508526 0.003396 0.003379 0.003043 0.002067 0.003957
C30 0.538707 0.436155 0.002559 0.004363 0.002164 0.001963 0.003036
7 8 9
A1 0.002092 0.003563 0.003225
A2 0.004501 0.006037 0.006052
A3 0.005641 0.002942 0.004209
A4 0.003709 0.003524 0.005095
A5 0.002680 0.001801 0.003575
A6 0.000666 0.001586 0.001716
A7 0.000834 0.001354 0.003427
A8 0.004297 0.006167 0.003379
A9 0.000437 0.000327 0.000386
A10 0.002804 0.003297 0.007882
A11 0.001132 0.000868 0.000726
A12 0.001422 0.000792 0.000950
A13 0.003638 0.002616 0.001683
A14 0.007167 0.004085 0.003967
A15 0.004474 0.004684 0.006669
A16 0.003159 0.002478 0.003934
A17 0.006042 0.004700 0.003658
A18 0.002977 0.005370 0.003257
A19 0.002657 0.003816 0.005871
A20 0.007452 0.004825 0.005917
A21 0.002791 0.003060 0.003581
A22 0.002961 0.002990 0.007053
A23 0.004245 0.003203 0.004931
A24 0.003993 0.003596 0.003122
A25 0.000532 0.000615 0.000808
A26 0.003647 0.003466 0.003756
A27 0.003434 0.003999 0.005346
A28 0.008143 0.006795 0.007443
A29 0.003435 0.002767 0.002855
A30 0.003459 0.002532 0.004657
C1 0.001172 0.003364 0.002430
C2 0.000039 0.000049 0.000017
C3 0.000619 0.000508 0.000673
C4 0.000186 0.000152 0.000147
C5 0.000832 0.000648 0.000583
C6 0.000064 0.000037 0.000075
C7 0.001104 0.000598 0.001033
C8 0.003467 0.005217 0.002390
C9 0.001598 0.001836 0.003645
C10 0.005399 0.002425 0.003270
C11 0.000005 0.000009 0.000004
C12 0.000048 0.000025 0.000020
C13 0.001419 0.001207 0.001624
C14 0.000552 0.000956 0.000562
C15 0.001425 0.001001 0.000692
C16 0.001232 0.001081 0.000720
C17 0.000095 0.000109 0.000193
C18 0.000747 0.000813 0.000696
C19 0.001093 0.002242 0.000775
C20 0.000095 0.000062 0.000049
C21 0.000590 0.000516 0.000699
C22 0.000527 0.000654 0.000579
C23 0.004038 0.002645 0.002153
C24 0.000200 0.000223 0.000418
C25 0.002238 0.001802 0.002347
C26 0.002130 0.001693 0.002294
C27 0.001799 0.001373 0.001018
C28 0.002181 0.002332 0.002612
C29 0.003460 0.003039 0.004942
C30 0.002930 0.004656 0.003468 ,
causal spurious
A1 0.950897 0.049103
A2 0.984687 0.015313
A3 0.983385 0.016615
A4 0.986925 0.013075
A5 0.992222 0.007778
A6 0.982788 0.017212
A7 0.978950 0.021050
A8 0.974077 0.025923
A9 0.997676 0.002324
A10 0.956897 0.043103
A11 0.988859 0.011141
A12 0.996152 0.003848
A13 0.986352 0.013648
A14 0.974675 0.025325
A15 0.976806 0.023194
A16 0.987922 0.012078
A17 0.983731 0.016269
A18 0.989103 0.010897
A19 0.950376 0.049624
A20 0.983602 0.016398
A21 0.989367 0.010633
A22 0.979020 0.020980
A23 0.976485 0.023515
A24 0.982626 0.017374
A25 0.995422 0.004578
A26 0.985762 0.014238
A27 0.994027 0.005973
A28 0.992273 0.007727
A29 0.981425 0.018575
A30 0.964233 0.035767
C1 0.988285 0.011715
C2 0.999802 0.000198
C3 0.991132 0.008868
C4 0.999409 0.000591
C5 0.990410 0.009590
C6 0.999145 0.000855
C7 0.993644 0.006356
C8 0.970939 0.029061
C9 0.990124 0.009876
C10 0.974697 0.025303
C11 0.999954 0.000046
C12 0.999792 0.000208
C13 0.991032 0.008968
C14 0.997400 0.002600
C15 0.988249 0.011751
C16 0.990461 0.009539
C17 0.999602 0.000398
C18 0.992919 0.007081
C19 0.991733 0.008267
C20 0.999453 0.000547
C21 0.994589 0.005411
C22 0.998474 0.001526
C23 0.984372 0.015628
C24 0.999443 0.000557
C25 0.993789 0.006211
C26 0.990490 0.009510
C27 0.993413 0.006587
C28 0.988770 0.011230
C29 0.970252 0.029748
C30 0.992094 0.007906)
model_2L.compute_information_flow(plot_info_flow=True, save_fig=True, save_dir=output_dir)
( 0 1 2 3 4 5 6 \
A1 0.630852 0.318585 0.003944 0.010081 0.008394 0.004038 0.005039
A2 0.764985 0.223592 0.000991 0.002227 0.001554 0.001667 0.001201
A3 0.724908 0.224540 0.003001 0.011070 0.007717 0.004274 0.007849
A4 0.531217 0.370330 0.006179 0.016631 0.018012 0.011826 0.013029
A5 0.584612 0.396728 0.001299 0.004230 0.002651 0.001875 0.002716
A6 0.522168 0.390127 0.005233 0.017971 0.012055 0.006570 0.014431
A7 0.483343 0.448730 0.005821 0.016985 0.010296 0.008155 0.004118
A8 0.562059 0.381275 0.006224 0.014725 0.005924 0.007489 0.005646
A9 0.579477 0.368723 0.006213 0.008844 0.007335 0.005332 0.003678
A10 0.520144 0.390575 0.006226 0.023117 0.007767 0.011165 0.011706
A11 0.661750 0.298141 0.003913 0.004610 0.006608 0.003019 0.004876
A12 0.673785 0.283051 0.002476 0.010025 0.007173 0.003379 0.007528
A13 0.630290 0.314376 0.003175 0.015003 0.005332 0.007056 0.005715
A14 0.554140 0.342883 0.009950 0.028887 0.012746 0.008813 0.010348
A15 0.605751 0.371448 0.002433 0.002343 0.004482 0.001948 0.003095
A16 0.544656 0.383367 0.005408 0.015162 0.008359 0.007309 0.007341
A17 0.569135 0.397980 0.003331 0.006159 0.004124 0.002287 0.007216
A18 0.459906 0.452811 0.007125 0.014549 0.014591 0.006700 0.017078
A19 0.512536 0.393885 0.005153 0.017646 0.014517 0.006019 0.014347
A20 0.646742 0.285145 0.005748 0.018098 0.010754 0.007316 0.004383
A21 0.526865 0.429176 0.003331 0.009318 0.007296 0.005465 0.005163
A22 0.522559 0.405968 0.003974 0.016192 0.017548 0.004371 0.006613
A23 0.651508 0.293734 0.003454 0.008671 0.006495 0.004420 0.012690
A24 0.608412 0.306895 0.005261 0.022785 0.012313 0.010387 0.007269
A25 0.586648 0.396831 0.001091 0.002881 0.003370 0.001791 0.002653
A26 0.581889 0.355270 0.005351 0.010787 0.009465 0.012801 0.007801
A27 0.516551 0.410190 0.005050 0.017993 0.008291 0.004955 0.007690
A28 0.599887 0.357194 0.003687 0.008658 0.003910 0.004613 0.007852
A29 0.507340 0.408165 0.004413 0.020353 0.016572 0.005516 0.007222
A30 0.567321 0.372151 0.005275 0.011789 0.009107 0.003837 0.014339
C1 0.643585 0.316641 0.003021 0.006518 0.007342 0.005766 0.003745
C2 0.541852 0.391603 0.004846 0.017364 0.007647 0.006649 0.006658
C3 0.438754 0.517998 0.002520 0.006490 0.005874 0.006534 0.004604
C4 0.568677 0.339401 0.004001 0.017661 0.009178 0.009823 0.016509
C5 0.642626 0.318047 0.003111 0.005106 0.007953 0.003993 0.007378
C6 0.631181 0.333015 0.002390 0.006908 0.003961 0.003829 0.004584
C7 0.491950 0.483831 0.001351 0.004075 0.003003 0.002280 0.005450
C8 0.474616 0.477119 0.003745 0.013395 0.009055 0.004028 0.005392
C9 0.448716 0.511961 0.002592 0.006751 0.007480 0.004423 0.004653
C10 0.544741 0.357364 0.012565 0.015990 0.009213 0.011656 0.017721
C11 0.499674 0.450371 0.003726 0.008411 0.005574 0.005063 0.005088
C12 0.518229 0.394939 0.005557 0.018574 0.011529 0.008770 0.012222
C13 0.520225 0.413864 0.004188 0.012555 0.008665 0.010756 0.006817
C14 0.487949 0.447849 0.006228 0.017722 0.008323 0.007822 0.009249
C15 0.638558 0.333838 0.001517 0.007267 0.003201 0.002974 0.002691
C16 0.689980 0.277872 0.002541 0.003580 0.004451 0.003287 0.004617
C17 0.579555 0.407178 0.000657 0.003442 0.001834 0.001403 0.001637
C18 0.703768 0.228489 0.005956 0.016919 0.007975 0.004958 0.009829
C19 0.781555 0.174173 0.003563 0.007725 0.006524 0.003274 0.006154
C20 0.738882 0.238279 0.002131 0.006289 0.003999 0.001629 0.002190
C21 0.516308 0.408200 0.005390 0.019042 0.013610 0.006517 0.007339
C22 0.501446 0.437590 0.003531 0.014603 0.008874 0.005957 0.008276
C23 0.559623 0.389726 0.004855 0.006558 0.007310 0.005886 0.005935
C24 0.535000 0.359075 0.008154 0.027785 0.015076 0.006721 0.022632
C25 0.722750 0.248418 0.002050 0.003719 0.005528 0.002696 0.003729
C26 0.680486 0.292610 0.002771 0.004564 0.005080 0.004100 0.001354
C27 0.497896 0.470187 0.002826 0.006844 0.003649 0.003358 0.005679
C28 0.354893 0.587241 0.006495 0.008543 0.007060 0.003655 0.006971
C29 0.654419 0.315111 0.002209 0.005479 0.003549 0.005255 0.004096
C30 0.504934 0.408905 0.005079 0.018343 0.012936 0.011011 0.012249
7 8 9
A1 0.009724 0.005484 0.003860
A2 0.001802 0.001313 0.000667
A3 0.008155 0.006020 0.002466
A4 0.013697 0.012241 0.006838
A5 0.003038 0.001220 0.001630
A6 0.016420 0.010187 0.004839
A7 0.008612 0.010807 0.003134
A8 0.006237 0.007994 0.002427
A9 0.009083 0.008639 0.002674
A10 0.015610 0.006687 0.007003
A11 0.008912 0.004713 0.003457
A12 0.005284 0.004206 0.003094
A13 0.006430 0.007646 0.004976
A14 0.014991 0.009476 0.007767
A15 0.003253 0.002883 0.002363
A16 0.019629 0.005578 0.003191
A17 0.004333 0.002290 0.003144
A18 0.011228 0.007747 0.008266
A19 0.019473 0.009583 0.006841
A20 0.008595 0.008630 0.004591
A21 0.006539 0.004365 0.002482
A22 0.010876 0.009824 0.002077
A23 0.011405 0.005136 0.002487
A24 0.013981 0.005481 0.007216
A25 0.002468 0.001164 0.001102
A26 0.004839 0.006492 0.005304
A27 0.012388 0.012208 0.004684
A28 0.004833 0.005649 0.003716
A29 0.016491 0.008608 0.005320
A30 0.005961 0.006901 0.003319
C1 0.006847 0.004628 0.001906
C2 0.009491 0.007138 0.006752
C3 0.009893 0.004796 0.002537
C4 0.024058 0.007688 0.003005
C5 0.004645 0.003625 0.003516
C6 0.004525 0.005740 0.003866
C7 0.003467 0.003187 0.001407
C8 0.006733 0.003926 0.001990
C9 0.003413 0.006241 0.003770
C10 0.010507 0.013290 0.006953
C11 0.009252 0.004718 0.008124
C12 0.014652 0.009050 0.006478
C13 0.012730 0.005277 0.004921
C14 0.004925 0.005170 0.004763
C15 0.003657 0.004484 0.001813
C16 0.008474 0.002281 0.002918
C17 0.002050 0.001318 0.000927
C18 0.007230 0.008139 0.006737
C19 0.006115 0.006883 0.004035
C20 0.003231 0.002517 0.000854
C21 0.013514 0.006236 0.003844
C22 0.010940 0.003807 0.004975
C23 0.009127 0.008344 0.002635
C24 0.013967 0.007350 0.004240
C25 0.005951 0.002420 0.002739
C26 0.004001 0.002438 0.002597
C27 0.004088 0.002651 0.002821
C28 0.010722 0.010474 0.003946
C29 0.005433 0.001959 0.002490
C30 0.013142 0.006200 0.007201 ,
causal spurious
A1 0.919637 0.080363
A2 0.981292 0.018708
A3 0.963490 0.036510
A4 0.942143 0.057857
A5 0.969169 0.030831
A6 0.920049 0.079951
A7 0.937307 0.062693
A8 0.912605 0.087395
A9 0.960553 0.039447
A10 0.927247 0.072753
A11 0.970916 0.029084
A12 0.951627 0.048373
A13 0.957703 0.042297
A14 0.913393 0.086607
A15 0.987047 0.012953
A16 0.939060 0.060940
A17 0.988850 0.011150
A18 0.945254 0.054746
A19 0.946158 0.053842
A20 0.917838 0.082162
A21 0.954182 0.045818
A22 0.949843 0.050157
A23 0.955464 0.044536
A24 0.956205 0.043795
A25 0.981477 0.018523
A26 0.954817 0.045183
A27 0.905737 0.094263
A28 0.984437 0.015563
A29 0.948635 0.051365
A30 0.952422 0.047578
C1 0.963725 0.036275
C2 0.911030 0.088970
C3 0.933730 0.066270
C4 0.955408 0.044592
C5 0.941797 0.058203
C6 0.954146 0.045854
C7 0.977829 0.022171
C8 0.944057 0.055943
C9 0.959542 0.040458
C10 0.920482 0.079518
C11 0.926628 0.073372
C12 0.900053 0.099947
C13 0.922272 0.077728
C14 0.927123 0.072877
C15 0.977257 0.022743
C16 0.972236 0.027764
C17 0.985493 0.014507
C18 0.962885 0.037115
C19 0.948411 0.051589
C20 0.980467 0.019533
C21 0.915580 0.084420
C22 0.941026 0.058974
C23 0.965293 0.034707
C24 0.963340 0.036660
C25 0.964938 0.035062
C26 0.976311 0.023689
C27 0.976079 0.023921
C28 0.968024 0.031976
C29 0.983531 0.016469
C30 0.949127 0.050873)
adata_protein = sc.AnnData(X=microb.T)
adata_protein.var_names = microb_df.index
adata_protein.obs['labels_ori'] = labels.values
adata_protein.obs['labels'] = labels.apply(lambda tg: 1 if tg == labels[0] else 0)
adata_protein.obsm['X_ct_1'] = model_1L.get_model_output()['latent'][:,:2]
adata_protein.obsm['X_ct_2'] = model_2L.get_model_output()['latent'][:,:2]
sc.pp.neighbors(adata_protein)
WARNING: You’re trying to run this on 74 dimensions of `.X`, if you really want this, set `use_rep='X'`.
Falling back to preprocessing with `sc.pp.pca` and default params.
sc.tl.umap(adata_protein)
sc.pl.umap(adata_protein, color='labels_ori', palette={"Autism": "#D73027", "Control": "#4575B4"}, show=False, size=600)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, 'umap.pdf'))
Calculate causal/non-causal features contribution to PC loadings#
from sklearn.decomposition import PCA
# X: shape (n_samples, n_genes)
pca = PCA(n_components=20)
pca.fit(adata_layer2.X)
loadings = pca.components_.T # shape: (n_genes, 2)
df = pd.DataFrame(loadings[:, :2], columns=['PC1', 'PC2'], index=adata_layer2.var_names)
df['PC1_abs'] = np.abs(df['PC1'])
df['PC2_abs'] = np.abs(df['PC2'])
top_PC1 = df.sort_values('PC1_abs', ascending=False).head(10)
top_PC2 = df.sort_values('PC2_abs', ascending=False).head(10)
adata_layer2pca = pca.transform(adata_layer2.X)
import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
from scipy.stats import mannwhitneyu
import numpy as np
sns.set(style="whitegrid", palette="pastel")
k = 5
df = pd.DataFrame(adata_layer2pca[:, :k], columns=[f'PC{i+1}' for i in range(k)])
df['Group'] = adata_layer2.obs['labels_ori'].values
n_cols = 2
n_rows = int(np.ceil(k / n_cols))
fig, axes = plt.subplots(n_rows, n_cols, figsize=(6*n_cols, 5*n_rows), sharey=True)
axes = axes.flatten()
palette = {"Autism": "#D73027", "Control": "#4575B4"}
for i, ax in enumerate(axes):
if i >= k:
ax.axis('off')
continue
pc_name = f'PC{i+1}'
sns.boxplot(
x='Group',
y=pc_name,
data=df,
ax=ax,
palette=palette,
width=0.6,
showfliers=False
)
sns.swarmplot(
x='Group',
y=pc_name,
data=df,
ax=ax,
color='black',
alpha=1,
size=3
)
group1 = df[df['Group'] == df['Group'].unique()[0]][pc_name]
group2 = df[df['Group'] == df['Group'].unique()[1]][pc_name]
_, p_value = mannwhitneyu(group1, group2)
y_max = df[pc_name].max()
h = y_max * 0.05
ax.plot([0, 0, 1, 1], [y_max+h, y_max+2*h, y_max+2*h, y_max+h], lw=1.5, c='k')
if p_value < 0.001:
symbol = '***'
elif p_value < 0.01:
symbol = '**'
elif p_value < 0.05:
symbol = '*'
else:
symbol = 'ns'
ax.text(0.5, y_max+2.5*h, symbol, ha='center', va='bottom', color='k')
ax.set_title(pc_name, fontweight='bold')
ax.set_xlabel('')
ax.tick_params(axis='x', rotation=45)
axes[0].set_ylabel('PC Value', fontweight='bold')
plt.suptitle('Comparison of Principal Components Between Groups', y=1.05, fontsize=14, fontweight='bold')
plt.tight_layout()
plt.savefig(output_dir+'compare_PC1-5.pdf')
plt.show()
def plot_cumulative_contributions(pca, k=2, figsize=(6, 4)):
loading_dict = {
f'PC{i+1}_loading': np.abs(pca.components_[i])
for i in range(k)
}
df = pd.DataFrame({
'gene': adata_layer2.var_names,
'group': ['causal' if g in df_layer2.head(10).index else 'non-causal' for g in adata_layer2.var_names],
**loading_dict
})
fig, axes = plt.subplots(1, k, figsize=(figsize[0]*k, figsize[1]))
if k == 1:
axes = [axes]
for i in range(k):
pc_col = f'PC{i+1}_loading'
curves = {}
for group in df['group'].unique():
subset = df[df['group'] == group].copy()
sorted_vals = np.sort(subset[pc_col].values)
cum = np.cumsum(sorted_vals)
curves[group] = cum
ax = axes[i]
for group, cum in curves.items():
x = np.arange(len(cum))
ax.plot(x, cum, label=group, linewidth=4)
ax.set_title(f"Cumulative contribution to PC{i+1}")
ax.set_xlabel("Number of genes (sorted)")
ax.set_ylabel("Cumulative contribution")
ax.set_ylim(-0.1, 2.1)
ax.legend()
ax.grid(True)
plt.tight_layout()
plt.savefig(output_dir + 'cum_PC.pdf')
plt.show()
plot_cumulative_contributions(pca, k=3)
Significance testing for candidate triggers (permutation)#
adata_layer2.var_names.get_indexer(df_layer2.head(10).index)
array([59, 39, 19, 63, 37, 0, 38, 49, 64, 30])
# ======================================================
# SHAP → 全局置换检验（Global Permutation）
# ======================================================
df_shap_global, _ = model_2L.get_up_significance(
method="SHAP",
test_mode="permutation",
perm_mode="global",
n_perm=50,
n_bg_samples=60,
use_signed=True,
fdr_correct=False,
show_progress=True,
random_state=0,
)
[Note] SHAP-based significance computed via permutation test (global mode, 50 permutations).
print("df_shap_global: ", df_shap_global.iloc[[59, 39, 19, 63, 37, 0, 38, 49, 64, 30],:])
df_shap_global: weight weight_signed pvalue
Faecalibacterium 0.230254 0.004772 0.039216
Blautia 0.103730 -0.000170 0.549020
Akkermansia 0.087225 -0.003126 0.588235
Subdoligranulum 0.055883 0.000332 0.176471
Agathobacter 0.055594 0.000536 0.039216
Bacteroides 0.039992 0.001156 0.196078
Anaerostipes 0.036346 -0.000194 0.568627
Roseburia 0.026183 -0.000416 0.137255
Catenibacterium 0.022002 0.000731 0.352941
Streptococcus 0.019853 0.000369 0.176471
In-silico perturbation of significant triggers#
adata_TF = adata_layer2.copy()
adata_down = X_down.copy()
adata_down.obsm['X_umap'] = adata_protein.obsm['X_umap']
dot_size = 800
pert_Gene = ['Faecalibacterium', 'Roseburia']
fold = [2, 2]
_ = pert_plot_stream(adata_TF, adata_down, model_2L, pert_Gene,n_neighbors=None,scale=30,pert_celltype=['Autism'],run_suggest_mass_thresholds=True,fold=fold,state_obs='labels_ori', dot_size=dot_size, min_mass=1.3,embedding_name='X_umap')
dot_size = 800
pert_Gene = ['Faecalibacterium']
fold = [5]
model_output_pert1 = pert_plot_stream(
adata_TF, adata_down, model_2L,
pert_Gene,
n_neighbors=None,
scale=30,
pert_celltype=['Autism', 'Control'],
run_suggest_mass_thresholds=False,
fold=fold,
state_obs='labels_ori',
dot_size=dot_size,
min_mass=1.9,
embedding_name='X_umap',
palette={"Autism": "#D73027", "Control": "#4575B4"},
save_dir=output_dir
)
