This repository hosts the Graph4ASD Challenge, a graph machine learning competition focused on Autism Spectrum Disorder (ASD) classification using resting-state fMRI functional connectivity data from the ABIDE dataset.
Participants must design and train Graph Neural Network (GNN) models to classify subjects as:
based on brain connectivity graphs.
Each subject is represented as a graph where:
Leaderboard scores are automatically updated.
The evaluation metric is Macro F1-Score:
\[\text{Macro F1} = \frac{F1_{ASD} + F1_{TC}}{2}\]Rankings are sorted by descending score.
Each sample corresponds to one subject’s brain graph.
[ G = (A, X) ]
Where:
Train a model that correctly predicts whether a subject:
using:
This is a graph-level classification task.
The dataset is derived from:
Autism Brain Imaging Data Exchange (ABIDE)
ABIDE aggregates anonymized rs-fMRI data from 17 international sites.
| Property | Value |
|---|---|
| Subjects | 1,009 |
| ASD | 516 (51.14%) |
| Controls | 493 (48.86%) |
| Brain atlas | Craddock 200 |
⚠️ For the challenge, a processed subset is provided, 484 samples for train and 153 for test. Participants must use data as-is, since original data has been exclusively processed for this challenge.
Each subject is stored as a graph:
All graphs share:
The dataset is hosted in folder data/public.
Train your model using:
data/public/adj_train.npy and data/public/node_features.npy. Eachdata/public/train_label.csvPredict labels for every graph in:
data/public/adj_test.npy and data/public/node_features_test.npyCreate a metadata.json that contains metadata about your submission:
{
"team": "example_team",
"run_id": "example_run_id",
"type": "human", // must be "human", "llm-only", or "human+llm"
"model": "GAT",
"notes": "Additional notes"
}
Create a CSV with columns id and y_pred (same format as data/public/sample_submission.csv):
id,y_pred
1,0
2,0
3,1
...
Encrypt your predictions.csv using: python extra/encrypt.py predictions.csv.
Then put predictions.csv.enc and metadata.json in the submissions/ folder.
Note: ❗ Do NOT upload the raw CSV. You need to submit an encrypted version of your predictions file to keep privacy.
Commit and push the predictions.csv.enc and metadata.json to the repository. Then, open Pull Request or push to the main branch. The automated pipeline will decrypt and score your submission to update the leaderboard.
⚠️ Only one submission per participant is allowed.
Cameron Craddock, Yassine Benhajali, Carlton Chu, Francois Chouinard, Alan Evans, András Jakab, Budhachandra Singh Khundrakpam, John David Lewis, Qingyang Li, Michael Milham, Chaogan Yan, Pierre Bellec (2013). The Neuro Bureau Preprocessing Initiative: open sharing of preprocessed neuroimaging data and derivatives. In Neuroinformatics 2013, Stockholm, Sweden.