By adjusting the sliders on the left, you can visually explore how each latent factor affects the ECG reconstruction. The slider values represent standard deviations from the mean. The black line shows the new ECG reconstruction based on the current slider values, while the grey line represents the baseline ECG at the mean latent factor values.

You can also investigate interactions between latent factors by adjusting multiple sliders simultaneously. To reset all sliders back to their mean values, click Reset at the bottom of the page.
Latent Factor
Click on the gene(s) of interest in the left table to see how common variants in loci linked to these genes affect the ECG, shown through traversals of the related latent factors. Each plot corresponds to the reconstruction of the median beat exploring the influence of all associated latent factors while maintaining others at the mean value. Red lines indicate negative deviations, while blue lines indicate positive deviations.
Latent Factor
Click on a latent factor to display the latent factor traversal in the top plot, the heritability attributed to that latent factor, and the Manhattan plot in the bottom plot. The latent factor traversal plot shows how varying that single factor, across a range from –3 to +3 standard deviations from the mean, affects the reconstruction of the median ECG beat while keeping all other factors fixed at their mean values; red lines represent negative deviations, while blue lines represent positive deviations.

The heritability estimate indicates how much of the variance in the ECG is explained by genetic factors linked to that latent factor. The Manhattan plot provides a genome-wide overview of genetic associations, where each point represents a genetic variant plotted by chromosomal position (x-axis) and strength of association (–log₁₀ p-value) with the selected latent factor (y-axis); prominent peaks highlight genomic regions or loci that are significantly associated with variation in the latent factor, pointing to potential key genes or regions influencing ECG traits.
Associated genes
Legend

About

Latent factors in machine learning models like variational autoencoders (VAEs) are essentially hidden variables that capture underlying patterns in ECG signals.

Unlike traditional features that are manually defined and recognized visually, latent factors are learned directly from the data. They represent variations and essential characteristics of the ECG, allowing the model to uncover deeper insights and patterns without relying on pre-set assumptions or biases.

This approach not only makes the analysis more transparent but also helps in discovering novel relationships within the data, potentially offering new perspectives on cardiac electrophysiology.


This web app is currently being built using Shiny

References

REF

Frequently Asked Questions (FAQ)