Trophic dynamics is one of the major regulators of fishery production in marine ecosystems, which is important for the implementation of ecosystem-based fisheries management. Based on data collected form bottom trawl surveys in Haizhou Bay and adjacent waters during autumn of 2011 and 2018, Delta-GAMMs (Delta-generalized additive mixed models) were constructed to evaluate the effects of biotic and abiotic factors on the predation of five key prey species (including Leptochela gracilis, Alpheus japonicus, Loligo spp., Larimichthys polyactis, and Oratos- quilla oratoria) in the Haizhou Bay. Percent frequency of occurrence and predation pressure index were used to identify their major predators. Variance inflation factor and full subsets regression were analyzed to quantiiy the degree of multicollinearity between these factors. The results showed that the occurrence frequency of keystone prey species in the stomach of predators ranged from 8.5% to 42.2%, and the weight percentage ranged from 4.2% to 40.9%. The average deviance explanation rate of' the "binomial" model was 16.1%, and the average deviance explanation rate of the "positive" model was 23.8%. Body length of predator, predator population density, and sea bottom temperature were important factors influencing prey-predator trophic interactions. Predator length was the most important factor, with feeding probability and weight percentage of keystone prey species all increasing with body length of predator. Feeding probability and weight percentage of key prey species decreased with predator population density. The effects of environmental factors such as sea bottom temperature, depth of water, latitude, and sea bottom salinity showed different trends, depending on the prey-predator assemblage. This study showed that the Delta-GAMMs was an effective method to explore the trophic interactions between prey and predators in marine ecosystems, and could provide a theoretical basis for the conservation and sustainable use of fisheries resources. [ABSTRACT FROM AUTHOR]