Prediction of the development of islet autoantibodies through integration of environmental, genetic, and metabolic markers

Abstract

Background: The Environmental Determinants of the Diabetes in the Young (TEDDY) study has prospectively followed, from birth, children at increased genetic risk of type 1 diabetes. TEDDY has collected heterogenous data longitudinally to gain insights into the environmental and biological mechanisms driving the progression to persistent islet autoantibodies. Methods: We developed a machine learning model to predict imminent transition to the development of persistent islet autoantibodies based on timevarying metabolomics data integrated with time-invariant risk factors (eg, gestational age). The machine learning was initiated with 221 potential features (85 genetic, 5 environmental, 131 metabolomic) and an ensemble-based feature evaluation was utilized to identify a small set of predictive features that can be interrogated to better understand the pathogenesis leading up to persistent islet autoimmunity. Results: The final integrative machine learning model included 42 disparate features, returning a cross-validated receiver operating characteristic area under the curve (AUC) of 0.74 and an AUC of 0.65 on an independent validation dataset. The model identified a principal set of 20 time-invariant markers, including 18 genetic markers (16 single nucleotide polymorphisms [SNPs] and two HLA-DR genotypes) and two demographic markers (gestational age and exposure to a prebiotic formula). Integration with the metabolome identified 22 supplemental metabolites and lipids, including adipic acid and ceramide d42:0, that predicted development of islet autoantibodies. Conclusions: The majority (86%) of metabolites that predicted development of islet autoantibodies belonged to three pathways: lipid oxidation, phospholipase A2 signaling, and pentose phosphate, suggesting that these metabolic processes may play a role in triggering islet autoimmunity.

Prediction of type 1 diabetes at birth: cord blood metabolites versus genetic risk score in the MoBa cohort

“…[46] Webb-Robertson et al, using the TEDDY cohort, reported that the top 42 features (of 221) gave an AUC of 0.65 for prediction of islet autoimmunity in the holdout validation set, appreciably lower than the AUC of 0.74 estimated with cross-validation of the training set. [45] As seen in our results, cross-validation is essential to avoid overfitting, even when using few features and a relatively large sample size (166 cases, 177 controls). Several factors influence the potential for overfitting and bias, but in general, the larger the number of predictors in relation to the number of subjects with outcome, and the more flexible modelling approach (allowing non-linearities and interactions), the larger the potential for overfitting.…”

“…[44] Webb-Robertson et al used a machine learning approach to predict islet autoimmunity using data from the TEDDY study (157 case-control pairs), reporting azelaic acid and adipic acid as important features. [45] Frohnert et al predicted seroconversion to islet autoimmunity in the DAISY study (22 cases and 25 controls) and reported 3-methyloxobutyrate (a precursor to valine for leucine synthesis) and pyroglutamatic acid (a derivative of glutamic acid) as features that were often selected by the algorithm used. [46] We did not find any strong associations between specific metabolites measured at birth and later type 1 diabetes.…”

Prediction of Type 1 Diabetes at Birth: Cord Blood Metabolites vs Genetic Risk Score in the Norwegian Mother, Father, and Child Cohort

“…Frohnert et al, starting with 1552 features (and 22 cases), reported an AUC of 0.92 for prediction of islet autoimmunity in the DAISY study, without a holdout validation set ( 47 ). Webb-Robertson et al, using the TEDDY cohort, reported that the top 42 features (of 221) gave an AUC of 0.65 for prediction of islet autoimmunity in the holdout validation set, appreciably lower than the AUC of 0.74 estimated with cross-validation of the training set ( 46 ). As seen in our results, cross-validation is essential to avoid overfitting, even when using few features and a relatively large sample size (166 cases, 177 controls).…”

“…Stanfill et al used classification algorithms to determine the most predictive features of islet autoimmunity in the TEDDY study (504 samples) and reported adipic acid, creatinine, and leucine as influential metabolites ( 45 ). Webb-Robertson et al used a machine learning approach to predict islet autoimmunity using data from the TEDDY study (157 case-control pairs), reporting azelaic acid and adipic acid as important features ( 46 ). Frohnert et al predicted seroconversion to islet autoimmunity in the DAISY study (22 cases and 25 controls) and reported 3-methyl-oxobutyrate (a precursor to valine for leucine synthesis) and pyroglutamate acid (a derivative of glutamic acid) as features that were often selected by the algorithm used ( 47 ).…”

Technological readiness and implementation of genomic‐driven precision medicine for complex diseases

“…More broadly, AI has been used in decision support systems for health care providers, for example by helping predict the onset of septic shock in intensive care patients [71]. AI is also showing promise for the prediction of early disease onset, for example by determining the probability of developing islet autoantibodies in type 1 DM [72], and for prognostication in those already diagnosed with disease, such as in the development of psychosis in patients with other high-risk psychiatric conditions [73].…”