From medical professionals with cutting-edge technology to simple smartwatches, everything is generating data on an unprecedented scale. Aggregation of electronic health records, medical imaging, diagnostic tests, genomic data, and even real-time measurements from smartwatches creates a wealth of data for researchers and clinicians to analyze. These diverse data streams often have unique and overlapping signals, even within the same organ system.
For example, in the cardiovascular system, an electrocardiogram (ECG) measures electrical activity in the heart, and a photoplethysmogram (PPG), common on smartwatches, tracks changes in blood volume. Co-analysis of these modalities can simultaneously assess both the electrical system of the heart and pump efficiency, providing a more complete picture of heart health. Integrating these physiological signatures with genetic information from large national biobanks allows for the identification of the genetic basis of the disease.
Our previous study, Regle, has been successful in genetic discovery using health data, but was designed for a single data type (i.e., a unimodal setting). Alternatively, analyzing each modality individually and trying to piece together the findings later (what we call U-Regle or Unimodal Regle) may not be the most efficient method. U-Regle may miss subtle sharing information between different modalities. Instead, I assumed that Joint Modeling these complementary data streams increases important biological signals, reduces noise and leads to more powerful genetic discoveries.
Here we present a recent paper titled “Using multimodal AI to improve genetic analysis of cardiovascular properties.” American Journal of Human Genetics. We have developed a multimodal version of Regle called M-Regle. This allows multiple types of clinical data to be analyzed at once. M-Regle generates reduced reconstruction errors, identify more genetic associations, and outperforms risk scores in predicting heart disease compared to its predecessor, U-Regle.
