Assessing ion channel blockade and electromechanical biomarkers' interrelations through a novel Multi-Channel Causal Variational Autoencoder

Knowing the impact of causal relationships between ion channelblockade and electromechanical biomarkers is essential to improve drug-induced torsades de pointes (TdP)-risk assessment. Apart from commonpurely electric torsadogenic indices, mechanical biomarkers may provideadditional proarrhythmic information, but the impact and interrelationshipsbetween those variables need to be assessed to guide feature selection forclassification methods. Variational Autoencoders (VAE) offer a reliableframework for learning disentangled representations from complex data distributions and can handle a variety of data types and structures. Causal discovery strives to reveal causal links between observed variables, therebyproviding a better understanding and insight into the phenomenon under investigation. Nevertheless, establishing causal relationships between heterogeneous multichannel observations is far from straightforward. Wepropose a novel VAE architecture, Multi-Channel Causal Variational Autoencoder (MC$^2$VAE), to identify mutual relationships between ion channel blockades, torsadogenic biomarkers, and electromechanicalbiomarkers, considered here as threedistinct channels, i.e. three distinctsources of information for drug-induced TdP risk. Our approach forcausal disentanglement from multi-channel data is designed to search for a linear causal structure between thegenerated latent variables, shifting the problem of causal discovery from aheterogeneous multichannel space to a compact lower-dimensional one,while encoding and decoding operations are proper to each modality to betteradapt to their own specificities. Conclusion: Our approach interestingly suggests the existence of hidden (latent) causal relationships between the threeconsidered sets of biomarkers, providing a rationale for including mechanicalbiomarkers in TdP-risk assessment approaches. Further, MC$^2$VAE is able toquantify the strengths of the identified causal relationships, opening up a viable avenue for actionable interventions on the established graph.