Encoding Domain Information with Sparse Priors for Inferring Explainable Latent Variables

TL;DR

This paper introduces spex-LVM, a sparse prior-based latent variable model that enhances interpretability by integrating domain knowledge, effectively uncovering meaningful biological factors in high-dimensional data like single-cell RNA-seq.

Contribution

The paper presents spex-LVM, a novel factorial latent variable model that incorporates sparse priors and domain knowledge to produce explainable and interpretable latent factors.

Findings

Robustly identifies relevant biological structures.

Distinguishes technical noise from true variation.

Adapts pathway annotations to specific datasets.

Abstract

Latent variable models are powerful statistical tools that can uncover relevant variation between patients or cells, by inferring unobserved hidden states from observable high-dimensional data. A major shortcoming of current methods, however, is their inability to learn sparse and interpretable hidden states. Additionally, in settings where partial knowledge on the latent structure of the data is readily available, a statistically sound integration of prior information into current methods is challenging. To address these issues, we propose spex-LVM, a factorial latent variable model with sparse priors to encourage the inference of explainable factors driven by domain-relevant information. spex-LVM utilizes existing knowledge of curated biomedical pathways to automatically assign annotated attributes to latent factors, yielding interpretable results tailored to the corresponding domain of interest. Evaluations on simulated and real single-cell RNA-seq datasets demonstrate that our model robustly identifies relevant structure in an inherently explainable manner, distinguishes technical noise from sources of biomedical variation, and provides dataset-specific adaptations of existing pathway annotations. Implementation is available at https://github.com/MLO-lab/spexlvm.