Extracting Genetically-Imputed Causal Features From ECG Data

Area of Science:

• Cardiology
• Genetics
• Artificial Intelligence

Background:

• Atrial fibrillation (AF) is a common arrhythmia linked to stroke, heart failure, and mortality.
• Electrocardiograms (ECGs) are crucial for AF diagnosis, recording heart's electrical activity.
• Deep learning and Mendelian randomization (MR) are emerging tools for AF prediction and causal inference.

Purpose of the Study:

• To apply and extend the DeepFEIVR method for identifying genetically imputed causal ECG features associated with AF.
• To investigate the causal role of ECG genetic components in AF development using the UK Biobank dataset.
• To enhance DeepFEIVR and DeepFEIVR-RI for handling numerous instrumental variables (IVs) and to visualize extracted causal features.

Main Methods:

• Application of DeepFEIVR and DeepFEIVR-RI (a variant with residual inclusion) to the UK Biobank dataset.
• Extension of DeepFEIVR and DeepFEIVR-RI to accommodate a large number of IVs.
• Utilized dnn-loc algorithm for visual examination of extracted ECG causal features.

Main Results:

• Statistically significant evidence (p < 10^-8) that genetic components in ECGs contribute to AF development.
• Demonstrated the efficacy of DeepFEIVR and DeepFEIVR-RI in identifying disease-associated causal features from high-dimensional data.
• Provided a comparative analysis of DeepFEIVR and DeepFEIVR-RI using various IVs.

Conclusions:

• Genetic factors influencing ECG characteristics play a significant causal role in the etiology of atrial fibrillation.
• The extended DeepFEIVR framework effectively identifies genetic ECG components associated with AF, advancing causal inference in cardiovascular research.
• Visualizing extracted causal features aids in understanding the underlying mechanisms of AF.