Extended Data Fig. 3: Comparisons in associations between microbial species and type 2 diabetes across different statistical models.
Meta-analyzed associations of individual microbial species with type 2 diabetes (T2D) phenotype from the ordinal (a) and binary (b) models. The ordinal model modeled the disease status as an ordinal variable (T2D, prediabetes, or controls) and used data from all the participants. The binary model modeled the disease status as a binary variable (T2D or controls) and used data from T2D patients and normoglycemic controls. The blue-to-red and purple-to-orange gradients represent the magnitude and direction of the associations as quantified by meta-analyzed beta coefficients from linear mixed models adjusted for age, sex, and body mass index (BMI) and further adjusted for metformin use in MaAsLin2. All the results were corrected for multiple hypothesis testing by controlling the false discovery rate (FDR) using the Benjamini–Hochberg method with a target rate of 0.10. All models included each participant’s identifier as random effects and simultaneously adjusted for covariables. (c) Comparisons in associations between microbial species and T2D between multivariate MaAsLin2 models with and without further adjustment for BMI and metformin use from the ordinal model. (d) Comparisons in associations between microbial species and T2D between multivariate MaAsLin2 models with and without further adjustment for BMI and metformin use from the binary model. Dots in the scatter plots in (c) and (d) represent meta-analyzed beta coefficients from linear mixed models adjusted for covariables in MaAsLin2. All the statistical tests were two-sided. A total of 8,117 metagenomes from 1,851 T2D patients, 2,770 individuals with prediabetes, and 2,277 normoglycemic controls were included in the analyses in (a), (b), (c), and (d). Abbreviations: BMI, body mass index; Con, control; metf, metformin use; insul, insulin use; T2D, type 2 diabetes.
