Neural networks such as variational autoencoders (VAE) perform dimensionality reduction for the visualization and analysis of genomic data, but are limited in their interpretability: it is unknown which data features are represented by each embedding dimension. We present siVAE, a VAE that is interpretable by design, thereby enhancing downstream analysis tasks. Through interpretation, siVAE also identifies gene modules and hubs without explicit gene network inference. We use siVAE to identify gene modules whose connectivity is associated with diverse phenotypes such as iPSC neuronal differentiation efficiency and dementia, showcasing the wide applicability of interpretable generative models for genomic data analysis.

Intratumoral heterogeneity is associated with aggressive tumor behavior, therapy resistance, and poor patient outcomes. Such heterogeneity is thought to be dynamic, shifting over periods of minutes to hours in response to signaling inputs from the tumor microenvironment. However, models of this process have been inferred from indirect or post-hoc measurements of cell state, leaving the temporal details of signaling-driven heterogeneity undefined. Here, we developed a live-cell model system in which microenvironment-driven signaling dynamics can be directly observed and linked to variation in gene expression. Our analysis reveals that paracrine signaling between two cell types is sufficient to drive continual diversification of gene expression programs. This diversification emerges from systems-level properties of the EGFR-RAS-ERK signaling cascade, including intracellular amplification of amphiregulin-mediated paracrine signals and differential kinetic filtering by target genes including Fra-1, c-Myc, and Egr1. Our data enable more precise modeling of paracrine-driven transcriptional variation as a generator of gene expression heterogeneity. A record of this paper's transparent peer review process is included in the Supplemental Information.