Table 1.
Formulation of the experimental diets (%).
Table 2.
Overview of the eight feeding trials (Tr_1-8) conducted to optimize a larval diet for Japanese eel, A.japonica.
Fig 1.
Standardized survival (upper panels) and mean total length (lower panels) z-scores for each diet formulation across four feeding trials (Tr_1–Tr_4).
Bars represent z-scores for each period (6–40 dph, 41–80 dph, and 6–80 dph), colored as indicated in the legend. Missing data (i.e., diets not tested in certain periods) are not shown.
Fig 2.
Pairwise scatterplot matrix of standardized survival and mean total length (z-scores) for Japanese eel larvae across three developmental periods (6–40 dph, 41–80 dph, and 6–80 dph), based on data from feeding trials Tr_1–Tr_4.
Lower panels show scatterplots for each variable pair with colored linear regression lines (solid) and their corresponding 95% confidence intervals (shaded). Diagonal panels display the histogram of each variable. Upper panels indicate Pearson’s correlation coefficients (r), with significance levels denoted as *p < 0.05, **p < 0.01, ***p < 0.001. Each point represents a single diet formulation.
Table 3.
Result of 10-fold cross-validation by GPR model fitted to data from Tr_1-4.
Fig 3.
Parallel coordinate plot of candidate diet formulations generated by BO for each period (6–40 dph and 41–80 dph) using data from Tr_1–Tr_4.
Each line represents the ingredient composition of a single candidate diet, and the vertical axis indicates the percentage of each ingredient (expressed as a proportion out of 100 for all seven ingredients). Gray lines show the distribution of all candidate formulations, while the four diets (Nos. 41, 42, 43, and 44) selected for further validation are highlighted by distinct colors and line types.
Fig 4.
Comparison of survival rate and growth among diet groups in validation trials (Tr_5-6).
(A) Survival rate and (B) mean TL from 6 to 100 dph, shown as line plots by diet group in each trial. Data points represent the mean values from replicate tanks for each diet at each sampling time point. (C) Box plots of TL at 100 dph for each diet group in Tr_5 and Tr_6. Boxes indicate the interquartile range, the horizontal line represents the median, and white diamonds indicate the mean. Different lowercase letters above the boxes denote significant differences among diets within each trial (Tukey-Kramer test, p < 0.05). Sample sizes (n) for each group are indicated above the x-axis. Note: In Tr_5, due to a tank outflow accident in one replicate of the No.41 → 44 group at 76 dph, only one tank was available for data collection after 80 dph in this group.
Fig 5.
Comparison of survival rate and growth among diet groups in final validation trials (Tr_7-8).
(A) Survival rate and (B) mean TL from 6 to 100 dph, shown as line plots by diet group in each trial. Data points represent the mean values from replicate tanks (n = 2) for each diet at each sampling time point. (C) Box plots of TL at 100 dph for each diet group in Tr_7 and Tr_8. Boxes indicate the interquartile range, the horizontal line represents the median, and white diamonds indicate the mean. Different lowercase letters above the boxes denote significant differences among diets within each trial (Tukey-Kramer test, p < 0.05). Sample sizes (n) for each group are indicated above the x-axis. Note: In Tr_7 and Tr_8, the stocking density was adjusted at 41 dph. Therefore, after 41 dph, the survival rate was calculated by multiplying the cumulative survival rate up to 40 dph by the interval survival rate after 41 dph, using the number of larvae stocked at 41 dph as the new denominator.
Table 4.
Variable importance of proximate composition components in the random forest model fitted to data from Tr_1-8.
Fig 6.
SHAP dependence plots illustrating the effects of dietary proximate composition on standardized survival rate (blue) and mean TL (red) of Japanese eel larvae in the early (6–40 dph, upper row) and late (41–80 dph, lower row) rearing periods.
Each panel shows the relationship between the value of a given dietary component (x-axis) and its SHAP value (y-axis) as predicted by RF regression models trained for each period and trait, using data from all eight feeding trials (Tr_1–8). Points represent individual diets; loess curves indicate local trends, with shaded areas denoting 95% confidence intervals.