Understanding the predictions made by GNN models can lead to significant breakthroughs in trusting these models to be used for safety critical healthcare applications. GNNs are excellent at modeling non euclidean data such as graphs, but many scientists refrain from using them because of their blackbox nature. This is due to the ambiguity that exists regarding the background mechanisms that lead to predictions. Human brain can be effectively modeled as a graph, which is formally known as the brain network which is in turn used for studying interactions between various brain regions when it is affected by some disorder. Our work aims to conduct a comparative analysis among the explanations generated by various modified GNN explainability models for a pre-trained GNN model over Autism Spectrum Disorder (ASD) fMRI dataset. It mainly comprises 2 phases. The first being training a GNN model tailor-made for brain networks which highlights salient ROIs by using functional and structural information from fMRI. Secondly, the explanation generator that uses GNN explainability methods is modified specifically to suit brain network data to generate explanations about the predictions by discovering neurological biomarkers. We evaluate the performance of the explanation generator by employing various evaluation metrics. The obtained results show great correspondence with the ground truth explanations regarding the brain regions that get affected by autism.