A Next-Generation Artificial Intelligence-Based Integrated Genetic-Epigenetic Prediction of 5-Year Risk for Coronary Heart Disease

Author(s): Meesha Dogan, Cardio Diagnostics, Coralville IA; Ron Simons, Steve Beach, University of Georgia, Athens GA; Amaury Lendasse, University of Houston, Houston TX; Rob Philibert, Cardio Diagnostics, Coralville IA.

Presented: American Heart Association Scientific Sessions 2018

Coronary heart disease (CHD) is a leading cause of death in the US. CHD-associated morbidity and mortality can be reduced if those at risk can be identified well in advance, to put in place prevention interventions prior to an acute coronary event. Limitations of current approaches call for additional efforts to improve risk predictions. We recently developed a Precision Medicine tool capable of predicting 5-year risk, that can also guide personalized prevention interventions and monitor changes in risk over time. It was developed using an application of AI, machine learning, in the Framingham Heart Study to mine integrated genetic-epigenetic biosignatures that capture the complex interplay between our genome and the environment in conferring risk for CHD. The training and test sets consisted of 1180 (19/695 females and 23/485 males developed symptomatic CHD within 5-years) and 524 (8/293 females and 12/231 males developed symptomatic CHD within 5-years) individuals, respectively, of Northern European ancestry. Non-linear data mining and modeling techniques were employed on the training set to identify prediction signatures from genome-wide SNP and DNA methylation (DNAm) data. We identified 14 DNAm and 10 SNPs from the training set (training bootstrap average sensitivity and specificity of 0.61 and 0.71, respectively) that predicted the 5-year risk in the test set with a sensitivity and specificity of 0.60 and 0.69, respectively. Performance was compared to that of the Framingham Risk Score (FRS) and the ASCVD Risk Estimator. The sensitivity and specificity in the test set were 0.05 and 0.99 for FRS, and 0.38 and 0.85 for ASCVD. Even with a small number of incident cases, our tool performed with superior sensitivity, which is vital because, while a false positive would result in additional testing, a false negative can be detrimental. Unlike genetics, because DNAm is dynamic, it can be mapped to actionable risk factors to guide personalized interventions and over time, re-assess risk and continuously monitor heart health. We demonstrate translational feasibility of this tool via next-gen digital PCR assays. Additional efforts are ongoing to optimize, validate and extend our tool to further improve performance in larger, more ethnically diverse cohorts.

The post A Next-Generation Artificial Intelligence-Based Integrated Genetic-Epigenetic Prediction of 5-Year Risk for Coronary Heart Disease appeared first on Cardio Diagnostics.

An Integrated Genetic-Epigenetic Tool for Predicting Atherothrombotic Brain Infarction

Author(s): Meesha Dogan and Rob Philibert, Cardio Diagnostics, Coralville IA.

Presented: American Heart Association Scientific Sessions 2018

Atherothrombotic Brain Infarction (ABI) is a common, but potentially preventable cause of death. Unfortunately, secondary to the complexity of the genetic and environmental architecture underlying vulnerability to ABI, sensitive and specific tools to assess risk for and guide treatment this form of stroke have not yet been developed. However, in recent work, we have shown that integrated genetic-epigenetic tools can assess and predict risk for CHD which suggests that a similar approach may work for predicting ABI. Using a related machine learning approach involving gradient boosting and recursive feature elimination with cross-validation, we analyzed the data from 1248 subjects, including 26 individuals with ABI, from Wave 8 of the Framingham Heart Study. Results were compared to those obtained from prediction incorporating only conventional risk factors for stroke such as cholesterol. We found that post-tuning, a panel of 32 markers that included 3 SNPs and 29 DNA methylation markers predicted ABI with average ROC AUC, sensitivity and specificity of ABI status prediction were 0.89+/-0.09, 0.67+/-0.18, and 0.98+/-0.01, respectively. In comparison, after tuning, the average ROC AUC, sensitivity, and specificity of ABI status prediction using the risk factors was only 0.67+/-0.22, 0.12+/-0.10, and 0.94+/-0.01, respectively. Overall, our integrated tool has a >50% improvement in our sensitivity, >20% improvement in ROC AUC, and ~4% improvement in specificity. These results suggest the possibility that Precision Medicine tools that use machine-based learning algorithms with both genetic and epigenetic data that can also be translated into next-generation digital PCR assays may be able to accurately predict the occurrence of ABI. However, in order to effectively construct and implement these tools, larger data sets from longitudinally informative cohorts for ABI as well as other forms of stroke will be necessary.

The post An Integrated Genetic-Epigenetic Tool for Predicting Atherothrombotic Brain Infarction appeared first on Cardio Diagnostics.

Precision Cardiovascular Medicine: A Rapid and Highly Sensitive Artificial Intelligence-Based Integrated Genetic-Epigenetic DNA Test for 3-Year Risk for Incident Coronary Heart Disease.

Author(s): Meesha Dogan, Cardio Diagnostics, Coralville IA; Stacey Knight, Intermountain Healthcare, Salt Lake City, UT; Amaury Lendasse, University of Houston, Houston TX; Kirk Knowlton, Intermountain Healthcare, Salt Lake City, UT; Rob Philibert, Cardio Diagnostics, Coralville IA.

Presented: American Heart Association Scientific Sessions 2019

Coronary heart disease (CHD) associated morbidity and mortality are largely preventable. Primary prevention of CHD includes the estimation of risk using a risk estimator as a basis to recommend treatments to patients. Commonly utilized risk estimators such as the Framingham Risk Score (FRS) and the Pooled Cohort Equation (PCE) have several limitations. As an alternative, we have developed and externally validated a novel and simpler DNA-based integrated genetic-epigenetic 3-year risk estimator for incident CHD that is more sensitive for both men and women. This tool, which couples digital PCR-based DNA testing from blood or saliva with artificial intelligence can be performed in less than 12 hours. It was developed using DNA methylation (DNAm) and SNP data from the Framingham Heart Study (FHS) Offspring cohort (n=1172 in training set and n=512 in test set) and externally validated in an Intermountain (IM) cohort (n=80 in validation set and n=79 in test set). Data mining, feature selection, model development and model tuning were performed on the FHS training set and validated on the IM validation set. The final prediction model (ensemble of SVM, Random Forest and Logistic Regression), which consisted of 6 loci (3 DNAm and 3 SNPs) was tested on the FHS and IM test sets. The FRS and PCE risk calculators were implemented on all FHS and IM cohort data. Prediction results stratified by gender are shown below. The superior sensitivity of our tool in males and especially females will ensure that more individuals at risk for incident CHD are identified well in advance to allow better clinical decision-making on primary prevention strategies. Also, because DNAm signatures are dynamic and map to actionable risk factors, they may be leveraged to personalize interventions and monitor risk. A clinically implementable version of this tool has been developed as part of its translation into a Laboratory Developed Test, and is being extended to include more ethnically diverse cohorts.

The post A Rapid and Highly Sensitive Artificial Intelligence-Based DNA Test for 3-Year Risk for Incident Coronary Heart Disease appeared first on Cardio Diagnostics.

A Next-Generation Integrated Genetic-Epigenetic Algorithm to Predict Coronary Heart Disease
Author(s): Meeshanthini V Dogan, Cardio Diagnostics, Coralville, IA; Isabella Grumbach, Jacob Michaelson, Robert Philibert, University of Iowa, Iowa City, IA.

Presented: American Heart Association Scientific Sessions 2017

Coronary heart disease (CHD) is responsible for over 300,000 deaths annually in the United States. An improved method for detecting CHD could have a substantial clinical impact given that sudden cardiac death is the initial presentation in 15% of patients with CHD. Building on the idea that systemic effects of CHD risk factors are a conglomeration of genetic and environmental factors, we used machine learning techniques and integrated genetic, epigenetic, and phenotype data from the Framingham Heart Study to build and test a classification model for CHD. The initial development of this prediction model included 1545 individuals (58/851 females and 115/694 males diagnosed with CHD) in the training set and 142 individuals (22/54 females and 49/88 males diagnosed with CHD) in the test set. An ensemble of Random Forest classifiers consisting of 4 DNA methylation (DNAm) sites, 2 SNPs, age, and gender was capable of predicting CHD status with a test sensitivity and specificity of 0.75 and 0.80, respectively. In contrast, an ensemble consisting of conventional CHD risk factors (age, gender, systolic blood pressure, total cholesterol, HDL cholesterol, diabetes, and smoking status) performed with a test sensitivity and specificity of 0.41 and 0.89, respectively. The 4 DNAm sites captured variance associated with age, gender, total cholesterol, smoking, and diabetes. Our integrated approach with superior minimization of false negatives may be a viable alternative for CHD diagnosis and may obviate the need for cumbersome testing procedures, the collection of considerable amounts of blood, and multiple lab tests. Conceivably, these will be greatly reduced by using a single genetic-epigenetic assay that uses a microgram or less of blood or saliva DNA. More importantly, the methylation at epigenetic loci with the high predictive value could be very useful in guiding therapeutic interventions, management of risk factors, and monitoring the efficacy of treatments and lifestyle modifications. Optimization of our current panel led to a better selection of biomarkers and improved testing metrics. We further demonstrated the translational feasibility of our method by developing an alpha biomarker panel.

The post A Next-Generation Integrated Genetic-Epigenetic Algorithm to Predict Coronary Heart Disease appeared first on Cardio Diagnostics.