VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI.

Code ↔ Paper

4 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 4 matches

[1] § Methods › Implementation Details ↔ ASD_classification_ABIDEI.py, lines 197–258 · score 0.68 · linear warmup cosine, annealing learning rate, warmup epochs, Adam, scheduler, VarCoNet
[2] § Methods › Implementation Details ↔ subject_fingerprinting.py, lines 90–227 · score 0.68 · linear warmup cosine, annealing learning rate, warmup epochs, Adam, scheduler, VarCoNet
[3] § Results › ASD Classification Performance ↔ result_scripts/display_results_ABIDE.py, lines 1132–1211 · score 0.64 · BAnD, CVFormer, DeepFMRI, GCDA, GCL, UCGL
[4] § Methods › VarCoNet › rs‐fMRI Encoder ↔ model_scripts/VarCoNet.py, lines 26–60 · score 0.55 · positional encodings, Transformer encoder, trainable, kernels, CNN, layer

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Python · 386 lines · 18 KB · MIT · 1 match

from torch.utils.data import DataLoader
import numpy as np
import torch
from utils import InfoNCE
from tqdm import tqdm
from torch.optim import Adam
from utils import DualBranchContrast
from pl_bolts.optimizers import LinearWarmupCosineAnnealingLR
from model_scripts.VarCoNet import VarCoNet
from utils import ABIDEDataset
import os
from model_scripts.classifier import LREvaluator
import pickle
from sklearn.model_selection import train_test_split, StratifiedKFold
from utils import augment, removeDuplicates, test_augment_overlap
import copy
import argparse
def train(x, encoder_model, contrast_model, optimizer):
encoder_model.train()
optimizer.zero_grad()
z1 = encoder_model(x[0])
z2 = encoder_model(x[1])
loss = contrast_model(z1, z2)
loss.backward()
optimizer.step()
return loss.item(), z1.shape[1]
def test(encoder_model, train_loader, val_loader, test_loader,
min_length, max_length, num_classes, device, num_epochs, lr):
encoder_model.eval()
with torch.no_grad():
outputs_train = []
y_train = []
for (x,y) in train_loader:
x = x.to(device)
y_train.append(y)
outputs_train.append(encoder_model(x))
outputs_train = torch.cat(outputs_train, dim=0).clone().detach()
y_train = torch.cat(y_train,dim=0).to(device)
outputs_val = []
y_val = []
for (x,y) in val_loader:
x = x.to(device)
y_val.append(y)
outputs_val.append(encoder_model(x))
outputs_val = torch.cat(outputs_val, dim=0).clone().detach()
y_val = torch.cat(y_val,dim=0).to(device)
outputs_test= []
y_test = []
for (x,y) in test_loader:
x = x.to(device)
y_test.append(y)
outputs_test.append(encoder_model(x))
outputs_test = torch.cat(outputs_test, dim=0).clone().detach()
y_test = torch.cat(y_test,dim=0).to(device)
result,linear_state_dict = LREvaluator(num_epochs=num_epochs,learning_rate=lr).evaluate(encoder_model, outputs_train, y_train, outputs_val, y_val, outputs_test, y_test, num_classes, device)
return result,linear_state_dict
def main(config):
path = config['path_data']
names = []
with open(os.path.join(path,'ABIDEI_nilearn_names.txt'), 'r') as f:
for line in f:
names.append(line.strip())
data_list = np.load(os.path.join(path,'ABIDEI_nilearn_' + config['atlas'] + '.npz'))
data = []
for key in data_list:
data.append(data_list[key])
names_unique, counts = np.unique(names, return_counts=True)
names_dupl = names_unique[counts > 1]
pos_duplicates = []
names_duplicate = []
for name in names_dupl:
temp = np.where(np.array(names) == name)[0]
for t in temp:
pos_duplicates.append(t)
names_duplicate.append(name)
names_unique = names_unique[counts == 1]
pos_unique = []
for name in names_unique:
pos_unique.append(np.where(np.array(names) == name)[0][0])
train_DATA = [data[i] for i in pos_duplicates]
data = [data[i] for i in pos_unique]
y = np.load(os.path.join(path,'ABIDEI_nilearn_classes.npy'))
Y_train = y[pos_duplicates]
y = y[pos_unique]
ext_test = list(range(51456,51494))
names_ext_test = []
for name in ext_test:
if 'sub-00'+str(name) in names_unique:
names_ext_test.append('sub-00'+str(name))
names = []
for name in names_unique:
if name not in names_ext_test:
names.append(name)
pos_ext_test = []
for name in names_ext_test:
pos_ext_test.append(np.where(np.array(names_unique) == name)[0][0])
ext_test_data = [data[i] for i in pos_ext_test]
y_ext_test = y[pos_ext_test]
pos = []
for name in names:
pos.append(np.where(np.array(names_unique) == name)[0][0])
data = [data[i] for i in pos]
y = y[pos]
device = torch.device(config['device']) if torch.cuda.is_available() else torch.device("cpu")
max_length = data[0].shape[0]
train_length_limits = [config['min_length'], max_length]
eval_epochs = list(range(1, config['epochs']+1))
model_config = config['model_config']
model_config['max_length'] = max_length
'''------------------------------------KFold CV------------------------------------'''
losses_all = []
test_result_all = []
min_val_loss_epochs = []
min_loss_epochs = []
names_train_all = []
names_val_all = []
names_test_all = []
for i in range(10):
skf = StratifiedKFold(n_splits=10, shuffle = True, random_state=42+i)
for j, (train_index, test_index) in enumerate(skf.split(data, y)):
train_data = [data[n] for n in train_index]
test_data = [data[n] for n in test_index]
y_train = y[train_index]
y_test = y[test_index]
names_train = [names[n] for n in train_index]
names_test = [names[n] for n in test_index]
train_data, val_data, y_train, y_val, train_idx, val_idx = train_test_split(train_data,
y_train,
np.arange(len(train_data)),
test_size=0.15,
random_state=42,
stratify=y_train)
names_val = [names_train[n] for n in val_idx]
names_train = [names_train[n] for n in train_idx]
train_data = train_DATA + train_data
y_train = np.concatenate((Y_train, y_train))
names_train = names_duplicate + names_train
train_dataset = ABIDEDataset(train_data, y_train)
train_loader = DataLoader(train_dataset, batch_size=config['batch_size'],
shuffle = config['shuffle'])
val_dataset = ABIDEDataset(val_data, y_val)
val_loader = DataLoader(val_dataset, batch_size=config['batch_size'])
test_dataset = ABIDEDataset(test_data, y_test)
test_loader = DataLoader(test_dataset, batch_size=config['batch_size'])
names_train_all.append(names_train)
names_val_all.append(names_val)
names_test_all.append(names_test)
roi_num = test_data[0].shape[1]
encoder_model = VarCoNet(model_config, roi_num).to(device)
contrast_model = DualBranchContrast(loss=InfoNCE(tau=config['tau']),mode='L2L').to(config['device'])
optimizer = Adam(encoder_model.parameters(), lr=config['lr'])
scheduler = LinearWarmupCosineAnnealingLR(
optimizer=optimizer,
warmup_start_lr = 1e-5,
warmup_epochs=config['warm_up_epochs'],
max_epochs=config['epochs'])
min_val_loss = 1000
test_result = []
losses = []
with tqdm(total=config['epochs'], desc='(T)') as pbar:
for epoch in range(1,config['epochs']+1):
total_loss = 0.0
batch_count = 0
for batch_idx, sample_inds in enumerate(train_loader.batch_sampler):
sample_inds = removeDuplicates(names_train,sample_inds)
batch_list = [train_data[i] for i in sample_inds]
batch_loader = DataLoader(batch_list, batch_size=len(batch_list), num_workers=4)
batch_data = next(iter(batch_loader))
batch_data = augment(batch_data,train_length_limits,device)
loss,input_dim = train(batch_data,encoder_model,contrast_model,
optimizer)
total_loss += loss
batch_count += 1
scheduler.step()
average_loss = total_loss / batch_count if batch_count > 0 else float('nan')
losses.append(average_loss)
pbar.set_postfix({'loss': average_loss})
pbar.update()
if epoch in eval_epochs:
res,linear_state_dict = test(encoder_model,train_loader,val_loader,test_loader,
config['min_length'],max_length,config['num_classes'],
config['device'],config['epochs_cls'],config['lr_cls'])
test_result.append(res)
if res['best_val_loss'] < min_val_loss:
min_val_loss = res['best_val_loss']
min_val_loss_model = copy.deepcopy(encoder_model.state_dict())
min_val_loss_classifier = copy.deepcopy(linear_state_dict)
min_val_loss_epoch = epoch
losses_all.append(losses)
test_result_all.append(test_result)
min_val_loss_epochs.append(min_val_loss_epoch)
if config['save_models']:
if not os.path.exists(os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet')):
os.makedirs(os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet'),exist_ok=True)
torch.save(min_val_loss_model, os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet','min_val_loss_model_rs' + str(i) + '_fold' + str(j) + '.pth'))
torch.save(min_val_loss_classifier, os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet','min_val_loss_classifier_rs' + str(i) + '_fold' + str(j) + '.pth'))
'''------------------------------------Ext. test------------------------------------'''
losses_all_ext = []
ext_test_result_all = []
min_val_loss_epochs_ext = []
min_loss_epochs_ext = []
names_train_ext_all = []
names_val_ext_all = []
for i in range(10):
train_data, val_data, y_train, y_val, train_idx, val_idx = train_test_split(data,
y,
np.arange(len(data)),
test_size=0.1,
random_state=42+i,
stratify=y)
names_val = [names[n] for n in val_idx]
names_train = [names[n] for n in train_idx]
train_data = train_DATA + train_data
y_train = np.concatenate((Y_train, y_train))
names_train = names_duplicate + names_train
train_dataset = ABIDEDataset(train_data, y_train)
train_loader = DataLoader(train_dataset, batch_size=config['batch_size'], shuffle = config['shuffle'])
val_dataset = ABIDEDataset(val_data, y_val)
val_loader = DataLoader(val_dataset, batch_size=config['batch_size'])
test_dataset = ABIDEDataset(ext_test_data, y_ext_test)
test_loader = DataLoader(test_dataset, batch_size=config['batch_size'])
names_train_ext_all.append(names_train)
names_val_ext_all.append(names_val)
roi_num = ext_test_data[0].shape[1]
encoder_model = VarCoNet(model_config, roi_num).to(device)
contrast_model = DualBranchContrast(loss=InfoNCE(tau=config['tau']),mode='L2L').to(config['device'])
optimizer = Adam(encoder_model.parameters(), lr=config['lr'])
scheduler = LinearWarmupCosineAnnealingLR(
optimizer=optimizer,
warmup_start_lr = 1e-5,
warmup_epochs=config['warm_up_epochs'],
max_epochs=config['epochs'])
min_val_loss = 1000
test_result = []
losses = []
with tqdm(total=config['epochs'], desc='(T)') as pbar:
for epoch in range(1,config['epochs']+1):
total_loss = 0.0
batch_count = 0
for batch_idx, sample_inds in enumerate(train_loader.batch_sampler):
sample_inds = removeDuplicates(names_train,sample_inds)
batch_list = [train_data[i] for i in sample_inds]
batch_loader = DataLoader(batch_list, batch_size=len(batch_list))
batch_data = next(iter(batch_loader))
batch_data = augment(batch_data,train_length_limits,device)
loss,input_dim = train(batch_data,encoder_model,contrast_model,
optimizer)
total_loss += loss
batch_count += 1
scheduler.step()
average_loss = total_loss / batch_count if batch_count > 0 else float('nan')
losses.append(average_loss)
pbar.set_postfix({'loss': average_loss})
pbar.update()
if epoch in eval_epochs:
res,linear_state_dict = test(encoder_model,train_loader,val_loader,test_loader,
config['min_length'],max_length,config['num_classes'],
config['device'],config['epochs_cls'],config['lr_cls'])
test_result.append(res)
if res['best_val_loss'] < min_val_loss:
min_val_loss = res['best_val_loss']
min_val_loss_model = copy.deepcopy(encoder_model.state_dict())
min_val_loss_classifier = copy.deepcopy(linear_state_dict)
min_val_loss_epoch = epoch
losses_all_ext.append(losses)
ext_test_result_all.append(test_result)
min_val_loss_epochs_ext.append(min_val_loss_epoch)
if config['save_models']:
if not os.path.exists(os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet')):
os.makedirs(os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet'),exist_ok=True)
torch.save(min_val_loss_model, os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet','min_val_loss_model_rs' + str(i) + '.pth'))
torch.save(min_val_loss_classifier, os.path.join(config['path_save'],'models_ABIDEI',config['atlas'],'VarCoNet','min_val_loss_classifier_rs' + str(i) + '.pth'))
results = {}
results['losses'] = losses_all
results['epoch_results'] = test_result_all
results['min_val_loss_epoch'] = min_val_loss_epochs
results['min_loss_epoch'] = min_loss_epochs
results['losses_ext'] = losses_all_ext
results['epoch_results_ext'] = ext_test_result_all
results['min_val_loss_epoch_ext'] = min_val_loss_epochs_ext
results['min_loss_epoch_ext'] = min_loss_epochs_ext
results['names_train'] = names_train_all
results['names_val'] = names_val_all
results['names_test'] = names_test_all
results['names_train_ext'] = names_train_ext_all
results['names_val_ext'] = names_val_ext_all
results['names_ext_test'] = names_ext_test
if config['save_results']:
if not os.path.exists(os.path.join(config['path_save'],'results_ABIDEI',config['atlas'])):
os.makedirs(os.path.join(config['path_save'],'results_ABIDEI',config['atlas']),exist_ok=True)
with open(os.path.join(config['path_save'],'results_ABIDEI',config['atlas'],'ABIDEI_VarCoNet_results.pkl'), 'wb') as f:
pickle.dump(results,f)
return results
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Run VarCoNet on ABIDE I for ASD classification')
parser.add_argument('--path_data', type=str,
help='Path to the dataset')
parser.add_argument('--path_save', type=str,
help='Path to save results')
parser.add_argument('--atlas', type=str, choices=['AICHA', 'AAL'], default='AICHA',
help='Atlas type to use')
parser.add_argument('--device', type=str, default='cuda:0',
help='Device to use for training')
parser.add_argument('--min_length', type=int, default=80,
help='Minimum length for augmentation')
parser.add_argument('--epochs', type=int, default=50,
help='Number of epochs')
parser.add_argument('--warm_up_epochs', type=int, default=10,
help='Number of warm up epochs for the lr scheduler')
parser.add_argument('--epochs_cls', type=int, default=150,
help='Number of epochs for the linear classification layer')
parser.add_argument('--lr_cls', type=float, default=5e-5,
help='Learning rate for the linear classification layer')
parser.add_argument('--num_classes', type=int, default=2,
help='Number of classes for the classification')
parser.add_argument('--save_models', action='store_true',
help='Flag to save trained models')
parser.add_argument('--save_results', action='store_true',
help='Flag to save results')
args = parser.parse_args()
config = {
'path_data': args.path_data,
'path_save': args.path_save,
'atlas': args.atlas,
'min_length': args.min_length,
'shuffle': True,
'epochs': args.epochs,
'warm_up_epochs': args.warm_up_epochs,
'epochs_cls': args.epochs_cls,
'lr_cls': args.lr_cls,
'num_classes': args.num_classes,
'save_models': args.save_models,
'save_results': args.save_results,
'device': args.device,
'model_config': {}
}
with open(f'best_params_VarCoNet_{config["atlas"]}.pkl', 'rb') as f:
best_params = pickle.load(f)
config['batch_size'] = best_params['batch_size']
config['tau'] = best_params['tau']
config['lr'] = best_params['lr']
config['model_config']['layers'] = best_params['layers']
config['model_config']['n_heads'] = best_params['n_heads']
config['model_config']['dim_feedforward'] = best_params['dim_feedforward']
results = main(config)

ASD_classification_ABIDEI.py at commit 545a462, under MIT · at the source

Overview

Authors: Charalampos Lamprou¹, Aamna Alshehhi^1,2, Leontios J. Hadjileontiadis^1,3, Mohamed L. Seghier^1,2

ORCID iDs: Aamna Alshehhi, Mohamed L. Seghier

Department of Biomedical Engineering and Biotechnology Khalifa University of Science and Technology Abu Dhabi UAE
Health Engineering Innovation Group (HEIG) Khalifa University of Science and Technology Abu Dhabi UAE
Department of Electrical and Computer Engineering Aristotle University of Thessaloniki Thessaloniki Greece

Institutions: Khalifa University of Science and Technology (United Arab Emirates); Aristotle University of Thessaloniki (Greece)

Journal: Human brain mapping, volume 47, issue 4, article e70469

Dates: received 22 September 2025; accepted 30 January 2026; published online 11 March 2026; in print March 2026

Type: Research article · Language: English

License: CC BY-NC

Identifiers: DOI 10.1002/hbm.70469 · PMID 41810518 · PMCID PMC12976809 · OpenAlex W7134922520

Open access: gold, a free copy (OpenAlex)

Status: code verified

Categories: fMRI (modality), human (organism), autism (population), methods / tools (subfield)

Methods: Spectral & time-frequency, Connectivity, Smoothing, state filtering, decompositions, Statistics, Machine learning, fMRI & imaging

Keywords: autism spectrum disorder classification, functional connectome, interindividual variability, resting‐state fMRI, self‐supervised learning, subject‐fingerprinting

MeSH: Autism Spectrum Disorder*, Brain*, Connectome*, Image Processing, Computer-Assisted*, Magnetic Resonance Imaging*, Nerve Net*, Supervised Machine Learning*, Convolutional Neural Networks, Humans (* major topic)

Topic: Functional Brain Connectivity Studies (Cognitive Neuroscience, Neuroscience), according to OpenAlex

Funding: Khalifa University of Science, Technology and Research (RC2‐2018‐022)

Citations: not cited yet (Europe PMC); 66 references in the paper

Abstract

Accounting for interindividual variability in brain function is key to precision medicine. Here, by considering functional interindividual variability as meaningful data rather than noise, we introduce VarCoNet, an enhanced self‐supervised framework for robust functional connectome (FC) extraction from resting‐state fMRI (rs‐fMRI) data. VarCoNet employs self‐supervised contrastive learning to exploit inherent functional interindividual variability, serving as a brain function encoder that generates FC embeddings readily applicable to downstream tasks even in the absence of labeled data. Contrastive learning is facilitated by a novel augmentation strategy based on segmenting rs‐fMRI signals. At its core, VarCoNet integrates a 1D‐convolutional neural network (CNN) with a Transformer encoder for advanced time‐series processing, enhanced with robust Bayesian hyperparameter optimization. Our VarCoNet framework is evaluated on two downstream tasks: (i) subject fingerprinting, using rs‐fMRI data from the Human Connectome Project (2117 recordings), and (ii) autism spectrum disorder (ASD) classification, using rs‐fMRI data from the Autism Brain Imaging Data Exchange (ABIDE) I (995 recordings) and II (730 recordings) datasets. Using different brain parcellations, our extensive testing against state‐of‐the‐art methods, including 13 deep learning methods, demonstrates VarCoNet's superiority, robustness, interpretability, and generalizability, achieving up to 98% subject fingerprinting accuracy and an area under the curve (AUC) of 72.6% for ASD classification. Overall, VarCoNet provides a versatile and robust framework for FC analysis in rs‐fMRI.

Reproduced under the paper's license (CC BY-NC), from the paper cited above.

Repositories

Its files are read in the Code ↔ Paper reader above, with 4 matches between paragraphs and lines of code.

LLNL/BAnD

License: none: the authors keep all their rights

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: bd9f19e767345cc5e6ed6a323d84e395d115e27c, 22 July 2021

Languages: Python (22), Shell (1)

Size: 25 files, 23 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README

Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: PyTorch (11 files), NumPy (4 files)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

24 files

band/__init__.py, Python, 1 line
band/models/TransformerBert.py, Python, 459 lines
band/models/TransformerFM.py, Python, 453 lines
band/models/TransformerFMAttention.py, Python, 465 lines
band/models/TransformerX.py, Python, 308 lines
band/models/__init__.py, Python, 1 line
band/models/models.py, Python, 707 lines
band/models/resnet.py, Python, 196 lines
band/models/resnet3d.py, Python, 557 lines
band/utilities/__init__.py, Python, 1 line
band/utilities/callbacks.py, Python, 492 lines
band/utilities/datasets.py, Python, 147 lines
band/utilities/fwriter.py, Python, 40 lines
band/utilities/glue.py, Python, 494 lines
band/utilities/misc.py, Python, 109 lines
band/utilities/scaffold.py, Python, 466 lines
band/utilities/tools.py, Python, 117 lines
band/utilities/transforms.py, Python, 640 lines
constants.py, Python, 6 lines
main.py, Python, 93 lines
main_dist.py, Python, 133 lines
main_dist_finetune.py, Python, 137 lines
start_dist.sh, Shell, 62 lines
README.md, Text, 15 lines

At the source: github.com/LLNL/BAnD

Wayfear/FBNETGEN

License: none: the authors keep all their rights

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 3ed6c813920966a40fca9ffca41f90a6384e333c, 18 May 2022

Languages: Python (19)

Size: 28 files, 19 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README

Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: NumPy (11 files), PyTorch (10 files), scikit-learn (5 files), pandas (4 files), Nilearn (2 files), SciPy (2 files)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

20 files

dataloader.py, Python, 204 lines
main.py, Python, 100 lines
model/GSL.py, Python, 434 lines
model/__init__.py, Python, 2 lines
model/cell.py, Python, 157 lines
model/model.py, Python, 450 lines
train.py, Python, 474 lines
util/FCNet/fc_net_label_generation.py, Python, 62 lines
util/FCNet/infer.py, Python, 116 lines
util/__init__.py, Python, 2 lines
util/abide/01-fetch_data.py, Python, 102 lines
util/abide/02-process_data.py, Python, 100 lines
util/abide/03-generate_abide_dataset.py, Python, 72 lines
util/abide/preprocess_data.py, Python, 272 lines
util/analysis/extract_info_from_log.py, Python, 32 lines
util/logger.py, Python, 19 lines
util/loss.py, Python, 55 lines
util/meter.py, Python, 70 lines
util/prepossess.py, Python, 82 lines
readme.md, Text, 68 lines

At the source: github.com/Wayfear/FBNETGEN

athms/learning-from-brains

License: none: the authors keep all their rights

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 126ab3c738ad81d50858ac90235820fae61b8b2e, 11 July 2023

Languages: Python (56)

Size: 85 files, 56 scripts

Software Heritage: archived

Found in: the text, “ASD Classification Baselines”

Holds: README, environment (Dockerfile, poetry.lock, pyproject.toml, requirements.txt), tests

Not found: license file, CITATION.cff, continuous integration, documentation

Tools: PyTorch (27 files), NumPy (18 files), pandas (11 files), Matplotlib (8 files), seaborn (8 files), Hugging Face Transformers (4 files), NiBabel (3 files), Nilearn (3 files), Pillow (1 file), scikit-learn (1 file), TemplateFlow (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

57 files

scripts/analyses/downstream_chance_accuracies.py, Python, 102 lines
scripts/analyses/fig2_upstream_data_overview.py, Python, 175 lines
scripts/analyses/fig3_hyperopt.py, Python, 209 lines
scripts/analyses/fig4_upstream_performance.py, Python, 134 lines
scripts/analyses/fig5_downstream_performance.py, Python, 150 lines
scripts/analyses/sfig1_training_curve_largest_sequence_BERT.py, Python, 98 lines
scripts/analyses/sfig2_eval_error_brain_maps.py, Python, 257 lines
scripts/analyses/sfig3_upstream_performance_pretrained_lms.py, Python, 133 lines
scripts/analyses/sfig4_5_downstream_learning_curves.py, Python, 226 lines
scripts/analyses/sfig6_downstream_performance_replication.py, Python, 55 lines
scripts/analyses/sfig7_downstream_learning_curves_replication.py, Python, 53 lines
scripts/dataprep/downstream/hcp_dataprep.py, Python, 297 lines
scripts/dataprep/downstream/hcp_to_bids.py, Python, 424 lines
scripts/dataprep/downstream/mdtb_dataprep.py, Python, 303 lines
scripts/dataprep/upstream/dataprep.py, Python, 211 lines
scripts/train.py, Python, 1,077 lines
src/__init__.py, Python, 21 lines
src/batcher/__init__.py, Python, 9 lines
src/batcher/base.py, Python, 304 lines
src/batcher/bert.py, Python, 432 lines
src/batcher/make.py, Python, 100 lines
src/decoder/__init__.py, Python, 17 lines
src/decoder/autoencoder.py, Python, 354 lines
src/decoder/bert.py, Python, 102 lines
src/decoder/gpt.py, Python, 165 lines
src/decoder/linear_baseline.py, Python, 82 lines
src/decoder/make.py, Python, 115 lines
src/embedder/__init__.py, Python, 19 lines
src/embedder/autoen.py, Python, 15 lines
src/embedder/base.py, Python, 286 lines
src/embedder/bert.py, Python, 225 lines
src/embedder/csm.py, Python, 240 lines
src/embedder/dummy.py, Python, 122 lines
src/embedder/make.py, Python, 150 lines
src/embedder/mnm.py, Python, 88 lines
src/embedder/msm.py, Python, 62 lines
src/embedder/netbert.py, Python, 225 lines
src/model.py, Python, 199 lines
src/preprocessor.py, Python, 514 lines
src/tools/__init__.py, Python, 14 lines
src/tools/brainmap.py, Python, 354 lines
src/tools/data.py, Python, 169 lines
src/tools/visualize.py, Python, 21 lines
src/tools/wandb.py, Python, 39 lines
src/trainer/__init__.py, Python, 10 lines
src/trainer/base.py, Python, 80 lines
src/trainer/make.py, Python, 220 lines
src/unembedder.py, Python, 135 lines
tests/__init__.py, Python, 202 lines
tests/test_adapt.py, Python, 79 lines
tests/test_checkpoints.py, Python, 201 lines
tests/test_embed_dims.py, Python, 86 lines
tests/test_freeze.py, Python, 131 lines
tests/test_from-pretrained.py, Python, 120 lines
tests/test_resume.py, Python, 73 lines
tests/test_train.py, Python, 86 lines
README.md, Text, 129 lines

At the source: github.com/athms/learning-from-brains

Wayfear/BrainNetworkTransformer

License: MIT

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 8a588aadad0166209269fa114e5df4e42209e207, 29 December 2022

Languages: Python (33)

Size: 58 files, 33 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README, license file

Not found: CITATION.cff, environment file, tests, continuous integration, documentation

Tools: PyTorch (22 files), NumPy (8 files), scikit-learn (4 files), pandas (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

35 files

source/__init__.py, Python, 1 line
source/__main__.py, Python, 46 lines
source/components/__init__.py, Python, 3 lines
source/components/logger.py, Python, 41 lines
source/components/lr_scheduler.py, Python, 55 lines
source/components/optimizer.py, Python, 79 lines
source/dataset/__init__.py, Python, 20 lines
source/dataset/abcd.py, Python, 58 lines
source/dataset/abide.py, Python, 28 lines
source/dataset/dataloader.py, Python, 109 lines
source/dataset/preprocess.py, Python, 24 lines
source/models/BNT/__init__.py, Python, 1 line
source/models/BNT/bnt.py, Python, 150 lines
source/models/BNT/components/__init__.py, Python, 1 line
source/models/BNT/components/transformer_encoder.py, Python, 25 lines
source/models/BNT/ptdec/__init__.py, Python, 1 line
source/models/BNT/ptdec/cluster.py, Python, 104 lines
source/models/BNT/ptdec/dec.py, Python, 90 lines
source/models/__init__.py, Python, 11 lines
source/models/base.py, Python, 15 lines
source/models/brainnetcnn.py, Python, 52 lines
source/models/fbnetgen.py, Python, 254 lines
source/models/transformer.py, Python, 65 lines
source/training/FBNettraining.py, Python, 104 lines
source/training/__init__.py, Python, 27 lines
source/training/training.py, Python, 199 lines
source/utils/__init__.py, Python, 4 lines
source/utils/accuracy.py, Python, 27 lines
source/utils/count_params.py, Python, 11 lines
source/utils/gumbel_softmax.py, Python, 36 lines
source/utils/hyperboloid.py, Python, 153 lines
source/utils/meter.py, Python, 75 lines
source/utils/prepossess.py, Python, 91 lines
LICENSE, License, 21 lines
readme.md, Text, 82 lines

At the source: github.com/Wayfear/BrainNetworkTransformer

qbmizsj/A-GCL

License: none: the authors keep all their rights

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 08339b719642a5d886688330a4b562a031961091, 3 November 2023

Languages: Python (27)

Size: 98 files, 27 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README

Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: PyTorch (22 files), PyTorch Geometric (16 files), NumPy (9 files), scikit-learn (6 files), SciPy (5 files), Matplotlib (1 file), pandas (1 file), seaborn (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

28 files

agcl_ABIDE.py, Python, 324 lines
agcl_ABIDE_queue.py, Python, 385 lines
datasets/__init__.py, Python, 1 line
datasets/abideDataset.py, Python, 128 lines
datasets/tu_dataset.py, Python, 216 lines
unsupervised/__init__.py, Python, 1 line
unsupervised/convs/GraphSAGE_conv.py, Python, 103 lines
unsupervised/convs/Graph_conv.py, Python, 97 lines
unsupervised/convs/__init__.py, Python, 7 lines
unsupervised/convs/gat_conv.py, Python, 294 lines
unsupervised/convs/gatv2_conv.py, Python, 171 lines
unsupervised/convs/gcn_conv.py, Python, 208 lines
unsupervised/convs/gine_conv.py, Python, 56 lines
unsupervised/convs/inits.py, Python, 56 lines
unsupervised/convs/wgin_conv.py, Python, 53 lines
unsupervised/embedding_evaluation.py, Python, 293 lines
unsupervised/encoder/__init__.py, Python, 3 lines
unsupervised/encoder/gatv2_encoder.py, Python, 116 lines
unsupervised/encoder/tu_encoder.py, Python, 107 lines
unsupervised/encoder/uni_encoder.py, Python, 161 lines
unsupervised/learning/__init__.py, Python, 4 lines
unsupervised/learning/classifier.py, Python, 81 lines
unsupervised/learning/ginfomax.py, Python, 204 lines
unsupervised/learning/ginfominmax.py, Python, 59 lines
unsupervised/learning/gsimclr.py, Python, 49 lines
unsupervised/utils.py, Python, 270 lines
unsupervised/view_learner.py, Python, 37 lines
README.md, Text, 294 lines

At the source: github.com/qbmizsj/A-GCL

icon-lab/BolT

License: GPL-3.0

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: f7c713f9eaf4c37855e84ba125ea2d2f14474fac, 21 June 2023

Languages: Python (47)

Size: 51 files, 47 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README, license file

Not found: CITATION.cff, environment file, tests, continuous integration, documentation

Tools: NumPy (27 files), PyTorch (17 files), pandas (10 files), scikit-learn (7 files), Matplotlib (3 files), Nilearn (3 files), seaborn (2 files), OpenCV (1 file), Pillow (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

49 files

Analysis/BrainMapping/__init__.py, Python, 1 line
Analysis/BrainMapping/brainMapper.py, Python, 128 lines
Analysis/BrainMapping/brainRegressor.py, Python, 257 lines
Analysis/BrainMapping/brainViz.py, Python, 157 lines
Analysis/BrainMapping/impTokenExtractor.py, Python, 175 lines
Analysis/BrainMapping/subjectReader.py, Python, 37 lines
Analysis/BrainMapping/vizSticher.py, Python, 35 lines
Analysis/TaskTimings/__init__.py, Python, 1 line
Analysis/TaskTimings/calculated/__init__.py, Python, 1 line
Analysis/TaskTimings/calculated/emotion.py, Python, 56 lines
Analysis/TaskTimings/calculated/gambling.py, Python, 67 lines
Analysis/TaskTimings/calculated/language.py, Python, 80 lines
Analysis/TaskTimings/calculated/motor.py, Python, 59 lines
Analysis/TaskTimings/calculated/relational.py, Python, 91 lines
Analysis/TaskTimings/calculated/social.py, Python, 60 lines
Analysis/TaskTimings/calculated/wm.py, Python, 59 lines
Analysis/TaskTimings/taskRelavance.py, Python, 336 lines
Analysis/__init__.py, Python, 1 line
Analysis/analysis_extractRawData.py, Python, 209 lines
Analysis/relevanceCalculator.py, Python, 129 lines
Dataset/DataLoaders/__init__.py, Python, 1 line
Dataset/DataLoaders/abide1Loader.py, Python, 46 lines
Dataset/DataLoaders/hcpRestLoader.py, Python, 42 lines
Dataset/DataLoaders/hcpTaskLoader.py, Python, 42 lines
Dataset/Prep/__init__.py, Python, 1 line
Dataset/Prep/prep_abide.py, Python, 57 lines
Dataset/Prep/prep_atlas.py, Python, 21 lines
Dataset/__init__.py, Python, 1 line
Dataset/dataset.py, Python, 124 lines
Dataset/datasetDetails.py, Python, 41 lines
Models/BolT/__init__.py, Python, 1 line
Models/BolT/bolT.py, Python, 171 lines
Models/BolT/bolTransformerBlock.py, Python, 459 lines
Models/BolT/hyperparams.py, Python, 41 lines
Models/BolT/model.py, Python, 108 lines
Models/BolT/run.py, Python, 156 lines
Models/BolT/util.py, Python, 29 lines
Models/SVM/__init__.py, Python, 1 line
Models/SVM/hyperparams.py, Python, 14 lines
Models/SVM/model.py, Python, 43 lines
Models/SVM/run.py, Python, 91 lines
Models/SVM/util.py, Python, 39 lines
Models/__init__.py, Python, 1 line
__init__.py, Python, 1 line
prep.py, Python, 22 lines
tester.py, Python, 93 lines
utils.py, Python, 229 lines
LICENSE, License, 674 lines
README.md, Text, 88 lines

At the source: github.com/icon-lab/BolT

SJYuCNEL/brain-and-Information-Bottleneck

License: none: the authors keep all their rights

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 35aa63da5464e1641eff2d2ea052a446a55e84ba, 6 November 2024

Languages: Python (11)

Size: 19 files, 11 scripts

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Holds: README

Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: PyTorch (11 files), PyTorch Geometric (8 files), NumPy (7 files), scikit-learn (3 files), SciPy (3 files), NetworkX (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

12 files

BrainIB_V1/data/create_data.py, Python, 80 lines
BrainIB_V1/graphCNNb.py, Python, 281 lines
BrainIB_V1/main.py, Python, 200 lines
BrainIB_V1/mlp.py, Python, 49 lines
BrainIB_V1/pre_subgraph.py, Python, 34 lines
BrainIB_V1/utilis.py, Python, 45 lines
BrainIB_V2/SGSIB/GNN.py, Python, 77 lines
BrainIB_V2/SGSIB/sub_graph_generator.py, Python, 75 lines
BrainIB_V2/SGSIB/utils.py, Python, 225 lines
BrainIB_V2/SGSIB_main.py, Python, 116 lines
BrainIB_V2/data/create_dataset.py, Python, 48 lines
README.md, Text, 328 lines

At the source: github.com/SJYuCNEL/brain-and-Information-Bottleneck

SJYuCNEL/brain-and-Information-Bottleneck.Graph

License: none: the authors keep all their rights

State: the link is dead, verified on 30 September 2026

Evidence: found in the paper

Software Heritage: not archived

Found in: the text, “ASD Classification Baselines”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Availability: 1 check, the latest on 30 September 2026: the link is dead

30 September 2026: the link is dead

At the source: github.com/SJYuCNEL/brain-and-Information-Bottleneck.Graph

CharLamp10/VarCoNet-V2

License: MIT

State: the link answers, verified on 30 September 2026

Evidence: files inventoried

Commit: 545a4624468e97b3771fc27c729cdc93761845d4, 7 November 2025

Languages: Python (45), R (3)

Size: 131 files, 48 scripts

Software Heritage: not archived

Found in: the text, “Implementation Details”

Holds: README, license file

Not found: CITATION.cff, environment file, tests, continuous integration, documentation

Tools: NumPy (40 files), PyTorch (26 files), scikit-learn (17 files), pandas (11 files), SciPy (6 files), Nilearn (4 files), data.table (3 files), Matplotlib (3 files), PyTorch Geometric (3 files), seaborn (3 files), ggplot2 (2 files), NiBabel (2 files), Plotly (1 file), statannotations (1 file)

Availability: 1 check, the latest on 30 September 2026: the link answers

30 September 2026: the link answers

50 files

ASD_classification_ABIDEI.py, Python, 386 lines, 1 match
ASD_classification_ABIDEII.py, Python, 174 lines
ASD_classification_ablations.py, Python, 750 lines
VarCoNet_BO.py, Python, 246 lines
competing_methods/AE_KSVD.py, Python, 327 lines
competing_methods/A_GCL.py, Python, 282 lines
competing_methods/BAnD.py, Python, 450 lines
competing_methods/BNT_gather_results.py, Python, 116 lines
competing_methods/BolT.py, Python, 441 lines
competing_methods/BrainIB.py, Python, 390 lines
competing_methods/CVFormer.py, Python, 507 lines
competing_methods/DeepFMRI.py, Python, 397 lines
competing_methods/FBNETGEN_gather_results.py, Python, 132 lines
competing_methods/GCDA.py, Python, 278 lines
competing_methods/LFB_gather_results.py, Python, 69 lines
competing_methods/PCC.py, Python, 166 lines
competing_methods/UCGL.py, Python, 261 lines
competing_methods/VAE_KSVD.py, Python, 334 lines
competing_methods/__init__.py, Python, 1 line
competing_methods/fbnetgen.py, Python, 279 lines
data_preparations/parcellate_ABIDEI.py, Python, 99 lines
data_preparations/parcellate_ABIDEII.py, Python, 92 lines
data_preparations/parcellate_HCP.py, Python, 54 lines
data_preparations/prepare_ABIDEII_data.py, Python, 88 lines
data_preparations/prepare_ABIDEI_data.py, Python, 113 lines
data_preparations/prepare_HCP_data.py, Python, 235 lines
data_preparations/prepare_data_BAnD.py, Python, 87 lines
data_preparations/rename_ABIDE_files.py, Python, 39 lines
data_preparations/subsampling_HCP_data.py, Python, 75 lines
model_scripts/VarCoNet.py, Python, 229 lines, 1 match
model_scripts/__init__.py, Python, 1 line
model_scripts/classifier.py, Python, 102 lines
model_scripts/competing_models.py, Python, 392 lines
model_scripts/evaluation.py, Python, 15 lines
predictions_stability.py, Python, 229 lines
result_scripts/ASD_feature_importance.py, Python, 44 lines
result_scripts/BO_plot.py, Python, 98 lines
result_scripts/ablation_plots.py, Python, 947 lines
result_scripts/display_results_ABIDE.py, Python, 1,275 lines, 1 match
result_scripts/intra_inter_subject_similarities.py, Python, 76 lines
result_scripts/plot_fingerprinting_results.py, Python, 117 lines
result_scripts/prepare_for_BrainNetViewer.R, R, 58 lines
result_scripts/prepare_for_rex.py, Python, 304 lines
result_scripts/rex_plots.R, R, 59 lines
result_scripts/rex_plots_2.R, R, 46 lines
subject_fingerprinting.py, Python, 289 lines, 1 match
subject_fingerprinting_ablations.py, Python, 453 lines
utils.py, Python, 894 lines
LICENSE, License, 21 lines
README.md, Text, 210 lines

At the source: github.com/CharLamp10/VarCoNet-V2

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

9 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
264 scripts, each with its path and the digest of its content;
4 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data Availability Statement

The data that support the findings of this study are openly available in VarCoNet‐V2 at https://github.com/CharLamp10/VarCoNet‐V2.

Reproduced under the paper's license (CC BY-NC), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 30 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 4 authors, 6 keywords, 9 MeSH terms, 1 funder, 51 references.

Cite

This paper

Lamprou, C., Alshehhi, A., Hadjileontiadis, L. J., & Seghier, M. L. (2026). VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI. Human brain mapping, 47(4), e70469. https://doi.org/10.1002/hbm.70469

BibTeX

@article{lamprou2026varconet, author = {Lamprou, Charalampos and Alshehhi, Aamna and Hadjileontiadis, Leontios J. and Seghier, Mohamed L.}, title = {{VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI}}, journal = {Human brain mapping}, year = {2026}, month = mar, volume = {47}, number = {4}, pages = {e70469}, publisher = {Wiley}, issn = {1065-9471}, doi = {10.1002/hbm.70469}, url = {https://doi.org/10.1002/hbm.70469}, pmid = {41810518}, pmcid = {PMC12976809} }

RIS

TY - JOUR AU - Lamprou, Charalampos AU - Alshehhi, Aamna AU - Hadjileontiadis, Leontios J. AU - Seghier, Mohamed L. TI - VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI T2 - Human brain mapping J2 - Hum Brain Mapp PY - 2026 DA - 2026/03/01 VL - 47 IS - 4 SP - e70469 SN - 1065-9471 PB - Wiley DO - 10.1002/hbm.70469 UR - https://doi.org/10.1002/hbm.70469 LA - en ER -

CSL-JSON

{ "id": "10.1002/hbm.70469", "type": "article-journal", "title": "VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI", "container-title": "Human brain mapping", "author": [ { "family": "Lamprou", "given": "Charalampos" }, { "family": "Alshehhi", "given": "Aamna" }, { "family": "Hadjileontiadis", "given": "Leontios J." }, { "family": "Seghier", "given": "Mohamed L." } ], "container-title-short": "Hum Brain Mapp", "volume": "47", "issue": "4", "page": "e70469", "DOI": "10.1002/hbm.70469", "PMID": "41810518", "PMCID": "PMC12976809", "ISSN": "1065-9471", "publisher": "Wiley", "URL": "https://doi.org/10.1002/hbm.70469", "language": "en", "issued": { "date-parts": [ [ 2026, 3, 1 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.

VarCoNet: A Variability-Aware Self-Supervised Framework for Functional Connectome Extraction From Resting-State fMRI.

Code ↔ Paper

Paper

The authors' code

Overview

Abstract

Repositories

LLNL/BAnD

Wayfear/FBNETGEN

athms/learning-from-brains

Wayfear/BrainNetworkTransformer

qbmizsj/A-GCL

icon-lab/BolT

SJYuCNEL/brain-and-Information-Bottleneck

SJYuCNEL/brain-and-Information-Bottleneck.Graph

CharLamp10/VarCoNet-V2

Tracing map

Data

Data Availability Statement

Versions

Version 1, 30 September 2026: the first record

Cite

This paper

Similar papers

Contribute

Claim this paper

Correct its record

Validate its tracing map

Add the badge to its README

Request its removal

Discussion, reproductions, activity