Fig. 4

Feature importance of CpG island methylation of TFAM from patients. a Methylation across the promoter CpG region of the TFAM gene was determined using overhang bisulfite sequencing. b Experimental paradigm for amplification of the bisulfite-converted DNA for 23 CpG sites proximal (Amplicon 1) and 19 CpG sites distal (Amplicon 2) to the TFAM start site. SHAP binary depiction with HbA1c of the interaction between c CpG24 methylation and HbA1c and d CpG29 methylation and HbA1c. e Not including HbA1c for binary classification representing the factors positively (red) and negatively (blue) impacting the construction of the model, with size of the bars depicting importance. f A modified T-Plot where the main effects of biomarkers on the prediction output are shown along the diagonal axis whereas interaction effects are shown off the diagonal. SHAP binary depiction without HbA1c of patient separation with the individual and correlated effects of CpG24 methylation and CpG29 methylation. g Using HbA1c for multiple classification representing the factors positively (red) and negatively (blue) impacting the construction of the model, with size of the bars depicting importance. h SHAP multiple classification depiction with HbA1c of the interaction between TFAM gene total methylation and HbA1c. SHAP values > 0.0 are diabetic (T2DM), SHAP values < 0.0 are non-diabetic (ND), SHAP values = 0 are either ND or T2DM without influence on the model. Groups are considered significantly different if P ≤ 0.05 = * compared to non-diabetic. All data are presented as the mean ± standard error of the mean (SEM). ND: non-diabetic; T2DM: type 2 diabetic; HbA1c: glycated hemoglobin; CpG: cytosine nucleotide followed by a guanine nucleotide; TFAM: transcription factor A, mitochondrial; binary: no diabetes and diabetes; multiple: no diabetes, prediabetes, and type 2 diabetes