graph-level anomaly detection (GLAD) aims to identify graphs that exhibit
notable dissimilarity compared to the majority in a collection. However,
current works primarily focus on evaluating graph-level abnormality while
failing to provide meaningful explanations for the predictions, w