Dataset

The dataset used in this competition is derived from the publicly available Protein-Protein Association (PPA) dataset, specifically the ogbg-ppa dataset. The original dataset consists of 37 classes and is widely used for graph classification tasks.

Dataset Selection and Modification

For this competition:

• Subset Selection:
  - 6 classes out of the 37 available classes were chosen at random.
  - 40% of the original dataset has been randomly selected.

• Noise Addition:
  - Different levels of symmetric and asymmetric noise were added to the labels.
  - Four distinct datasets were created, each incorporating:
  - Different percentages of noise in the labels, or
  - Different types of noise (symmetric vs. asymmetric).

Dataset Structure

The dataset can be downloaded here.

Folder Organization

Once downloaded, the dataset folder contains four subfolders named A, B, C, and D:

• Each subfolder corresponds to a dataset generated using the procedure described above.

• Inside each folder, you will find two files:
  - train.json.gz: Training set (with labels) for model training.
  - test.json.gz: Test set (without labels) for generating predictions.