Rapid classification and screening of alkali tolerance in Eriocheir sinensis based on CART decision tree

Saline-alkali water is a huge potential aquaculture resource in China. However, saline-alkali stress can cause slow growth and low survival rate of cultured animals, which has severely restricted the aquaculture and promotion of Eriocheir sinensis and other saline-alkali tolerant species. In order to establish a rapid assessment and screening method for the ability to tolerate alkali stress of E. sinensis and address the limitation of traditional laboratory alkali stress experiments, such as time and labor consumption, strong destructiveness and difficulty in large-scale application. In this study,204 female two-year-old E. sinensis cultured in Shandong area were selected as the research objects, A 60 mmol/L NaHCO₃ alkali stress experiment was conducted, and three phenotypic traits including carapace length (CL), carapace width (CW) and body weight (BW) were measured. Three trait ratio parameters obtained by standardization, namely carapace length/carapace width (X₁), body weight/carapace length (X₂) and body weight/carapace width (X₃), were used as input features, with the alkali-tolerant survival grade under alkali stress as the output target. A CART decision tree classification and evaluation model for alkali tolerance was constructed by IBM SPSS22.0 software.10-fold cross-validation was used to verify the model performance, and indicators such as accuracy, Kappa coefficient and precision were calculated based on the confusion matrix to evaluate the classification effect of the model. The results showed that there were significant differences in phenotypic indicators and trait ratio parameters among groups with different alkali-tolerant grades: X₁ of the strong alkali-tolerant group was significantly lower than that of the weak alkali-tolerant group, while X₂ and X₃ were significantly higher than those of the weak alkali-tolerant group. The survival time was significantly negatively correlated with X₁ (R=−0.184), and significantly positively correlated with X₂ and X₃ (R=0.177, R=0.163), but the absolute values of all correlation coefficients were lower than 0.2, indicating a weak linear correlation between a single indicator and alkali tolerance. The overall accuracy of the constructed CART decision tree model reached 60.78% with a Kappa coefficient of 0.41, and the precision of the model for identifying strong alkali-tolerant individuals was as high as 82.14%, which could meet the accuracy requirements of actual breeding screening. Three screening rules for strong alkali-tolerant individuals only relying on X₁ and X₂ were extracted from the model, corresponding to three types of strong alkali-tolerant individuals: "broad-carapace basic type" (X₁≤0.885 and X₂≤1.914, confidence level 76.92%), "broad-carapace transitional type" (0.892<X₁≤0.904 and 1.629＜X₂≤1.914, confidence level 83.33%) and "body weight compensatory type" (X₁>0.915 and 2.022＜X₂≤2.106, confidence level 88.89%), thus forming a comprehensive and high-confidence screening system. The validation experiment showed that the conforming screening rule group (73.33%±5.77%) had a significantly higher survival rate than the non-conforming group (26.67%±5.77%). further confirming the reliability and effectiveness of the model. In conclusion, the CART decision tree model constructed in this study can realize rapid non-destructive screening of strong alkali-tolerant E. sinensis individuals only through conventional morphological measurement, with the advantages of simple operation and strong on-site applicability. It provides a practical tool for the directional breeding of E. sinensis in saline-alkali water culture, and also a methodological reference for the rapid assessment of stress resistance traits of other aquatic animals.