Add to Chrome
Add to Firefox
Add to EdgeEncoding Domain Knowledge in Multi-view Latent Variable Models: A Bayesian Approach with Structured Sparsity
Apr 13, 2022
Many real-world systems are described not only by data from a single source but via multiple data views. For example, in genomic medicine, a patient can be described by data from different molecular layers. This raises the need for multi-view models that are able to disentangle variation within and across data views in an interpretable manner. Latent variable models with structured sparsity are a commonly used tool to address this modeling task but interpretability is cumbersome since it requires a direct inspection and interpretation of each factor via a specialized domain expert. Here, we propose MuVI, a novel approach for domain-informed multi-view latent variable models, facilitating the analysis of multi-view data in an inherently explainable manner. We demonstrate that our model (i) is able to integrate noisy domain expertise in form of feature sets, (ii) is robust to noise in the encoded domain knowledge, (iii) results in identifiable factors and (iv) is able to infer interpretable and biologically meaningful axes of variation in a real-world multi-view dataset of cancer patients.
