Objective This study aimed to screen key sugar metabolism indicators for evaluating sweetness quality in vegetable soybean during the early seed filling stage, thereby establishing an evaluation system and clarifying its physiological basis.

Method Nine vegetable soybean varieties predominantly cultivated in the Huaibei region of Jiangsu Province were used as materials. Twenty sugar metabolism-related indicators in leaves, podshells, seed coats, and embryos were measured at the R5.5 stage (characterized by flat fresh seeds that had not yet fully filled the pods), and sweetness sensory scores were evaluated at the R6.2 stage (when fresh seeds fully filled the pods and were plump). A sweetness prediction model was constructed using correlation analysis and stepwise regression, and its accuracy was validated. Furthermore, path analysis and dynamic changes analysis were employed to elucidate the regulatory characteristics of key indicators on sweetness.

Result The prediction model for seed sweetness established based on sugar metabolism indicators in different organs at the R5.5 stage, was: Sweetness value=0.078+0.029×Leaf sucrose phosphate synthase (SPS) activity-0.162×Podshell starch content-0.051×Podshell sucrose synthase (SuSy) activity+0.049×Seed coat cell wall-bound invertase (CWIN) activity+0.081×Embryo SPS activity. The coefficient of determination (R²) of the model was 0.9660, and the predicted values fitted well with the measured values (R²=0.9657). Path analysis revealed that embryo SPS activity (0.4926), leaf SPS activity (0.3652), and seed coat CWIN activity (0.2708) exhibited the strongest positive direct effects on sweetness, constituting a positive "source-flow-sink" driving pathway. Conversely, podshell starch content (−0.4189) and podshell SuSy activity (−0.2915) showed negative direct effects, acting as regulators of carbon flux distribution. Dynamic analysis indicated that embryo SPS activity, seed coat CWIN activity, and embryo sucrose content peaked synchronously at the R6.2 stage, while podshell starch content and podshell SuSy activity decreased continuously throughout seed filling, showing a dynamic correlation with embryo sucrose accumulation.

Conclusion This study constructed a predictive model for sweet taste quality based on multi-organ sugar metabolism indicators at the R5.5 stage. Through validation with an independent sample, it was confirmed that the model can accurately predict the sweet taste quality of seeds at the R6.2 stage. This method shifts the evaluation time forward to the early stage of seed filling, providing technical support for addressing the challenges of low breeding efficiency and inefficient production-marketing coordination in vegetable soybean. The study clarifies the roles of "source-flow-sink" synergistic driving and podshell starch degradation in sweet taste formation, laying a physiological foundation for the application of this technology.