This systematic review by Chan et al. examines the usage of graph neural network (GNN) models, trained on fMRI data, on finding biomarkers for psychiatric disorders. While previous reviews have written about the usage of fMRI data-trained GNN models, they do not go in depth on GNN use for biomarker discovery or extensively analyze GNN models specialized in processing fMRI data. Within the databases PubMed and Scopus, 65 studies - analyzing fMRI data-trained GNN models on attention-deficit hyperactive disorder (ADHD), autism spectrum disorder (ASD), major depressive disorder (MDD), and schizophrenia (SZ) - were included in the analysis for this review. From the results, it was found that while there were common, broad regions of the brain discovered to be associated with each disorder, specific biomarkers were not as consistent, and further studies would be needed to improve the anatomical characterization of each condition. Furthermore, many studies did not explicitly state how confounding variables - such as age and sex - were considered, which may have inflated model accuracy. Additionally, most studies utilized data consortiums to obtain training data, which pool multiple datasets together to increase the sample size of the training data. While this lessens the issue of small sample sizes, data consortium data are often biased toward their largest dataset included. Finally, standardization must be done for how the GNN models are trained, the criteria on how biomarkers are discovered, and the means of reporting biomarkers, as variation in these aspects hinders reproducibility of the studies.