Traditional approaches for studying potential interactions in marine ecosystems often struggle to fully capture all taxa in a community, especially rare species. This issue is particularly challenging in coastal waters with high biodiversity and spatiotemporal dynamics. In this study, we employed environmental DNA (eDNA) metabarcoding, utilizing multiple marker genes, to comprehensively investigate interspecific interactions across various domains in the subtropical coastal waters of Hong Kong. The southern and eastern regions of Hong Kong waters exhibit distinct environmental seasonality, and our investigation focused on comparing the potential interaction networks and the keystone taxa between these two regions. The putative species interaction networks across various groups (i.e., bacteria, protists, and metazoans) were revealed by using weighted correlation network analysis (WGCNA). Our results showed that primary consumers, mainly dinoflagellates and ciliates, were the dominant actors within the interaction networks, although their distributions varied between the two regions. Bacterial taxa from the Pseudomonadota groups primarily constituted saprobes in the southern region, while exhibiting an even distribution in the eastern region. The interaction network in the southern region was larger but less stable compared to the eastern region. This could be attributed to the stronger responses of keystone taxa to environmental variations and the relatively higher number of connectors (e.g., Akashiwo and Protoperidinium within Dinophyceae) in the eastern region. Our findings highlight the versatility of eDNA metabarcoding for studying potential species interactions, providing critical insights into ecosystem structure and stability, and offering suggestions for marine biodiversity conservation.