Fig. 2: Selected features for identifying renal dysfunction in DM patients, namely DKD (Model 2).

a Venn diagram of AI-selected features in Model 2. b Box plot of AI-selected features of metabolites in Model 2. The statistical analysis with p-values was performed by ANOVA for significant metabolites in the four groups. The t-test was used for multiple comparisons within the four groups test. Box plot: Box plot includes a box and a set of whiskers. The lower line of the box is represented as Q1 (25th percentile). The upper line of the box is represented as Q3 (75th percentile). The middle and bold line in the box is represented as median. In general, the boundary of the lower and upper whiskers is 1.5 interquartile ranges (IQR, IQR = Q3 − Q1) below the Q1 and 1.5 IQR above the Q3. The extreme values outside this boundary are considered as outliers and plotted as black dots. If all data points are between Q1 − 1.5 x IQR and Q3 + 1.5 x IQR, the boundary of the lower and upper whiskers should be minimum and maximum of the data. The error bars in this figure represent the lower and the upper whiskers defined above. c Pie charts indicating the genotype frequencies of SNPs using the SNP datasets obtained from the subjects. Number sign indicates that the signal from the SNP array was lower than the calling rate. The χ2 test was used for comparisons of genotype frequencies within the four groups. d Adjusted odds ratios of factors in backward logistic regression procedure associated with the occurrence of CKD among DM patients. The Wald test was used to construct 95% confidence interval (CI) and test the significance of adjusted odds ratios of risk factors. The error bars represent the lower bound and the upper bound of adjusted odds ratio of 95% confidence interval. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. ALDH1L1 aldehyde dehydrogenase 1 family member L1, LY6D lymphocyte antigen 6 family member D, PCDH9 protocadherin 9.