English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Latent variable models such as the Variational Auto-Encoder (VAE) have becomea go-to tool for analyzing biological data, especially in the field ofsingle-cell genomics. One remaining challenge is the interpretability of latentvariables as biological processes that define a cell's identity. Outside ofbiological applications, this problem is commonly referred to as learningdisentangled representations. Although several disentanglement-promotingvariants of the VAE were introduced, and applied to single-cell genomics data,this task has been shown to be infeasible from independent and identicallydistributed measurements, without additional structure. Instead, recent methodspropose to leverage non-stationary data, as well as the sparse mechanism shiftassumption in order to learn disentangled representations with a causalsemantic. Here, we extend the application of these methodological advances tothe analysis of single-cell genomics data with genetic or chemicalperturbations. More precisely, we propose a deep generative model ofsingle-cell gene expression data for which each perturbation is treated as astochastic intervention targeting an unknown, but sparse, subset of latentvariables. We benchmark these methods on simulated single-cell data to evaluatetheir performance at latent units recovery, causal target identification andout-of-domain generalization. Finally, we apply those approaches to tworeal-world large-scale gene perturbation data sets and find that models thatexploit the sparse mechanism shift hypothesis surpass contemporary methods on atransfer learning task. We implement our new model and benchmarks using thescvi-tools library, and release it as open-source software athttps://github.com/Genentech/sVAE.

Learning Causal Representations of Single Cells via Sparse Mechanism Shift Modeling