<p>Mediation analysis serves as a foundational statistical approach to comprehending the impact of exposure on an outcome. In this study, we investigate group inference for high-dimensional mediation models, examining mediators within a specified group either jointly or individually, allowing the number of mediators within the group to diverge. For both situations, we construct suitable test statistics and establish their asymptotic distributions. Extensive numerical studies demonstrate the superiority of our proposed methods over recent representative approaches. Our procedure can control the type I error well and exhibit the highest power. We also apply our methods to analyse how Deoxyribonucleic acid (DNA) methylation operates in the regulation of human stress reactivity impacted by childhood trauma. We have pinpointed seven key biological process groups, with the top five significant groups-axon development, neuron projection regeneration, positive regulation of the Mitogen-activated protein kinases (MAPK) cascade, regulation of neuron projection development, and axonogenesis-playing a collective role in nurturing nerve cell growth, development, and signal transmission.</p>