ResolVI models measurement errors to fix single-cell RNA data, cutting false signals from 14.8% to under 0.01%

bravo_abad · x · 2026-09-26

Ergen and Yosef present ResolVI, a variational autoencoder that corrects measurement artifacts in single-cell RNA sequencing.

The problem: RNA molecules get assigned to the wrong cell, so cells appear to express genes from their neighbors, creating spurious cell states.
The approach: model the measured signal as a mixture of the cell's own RNA, nearby-cell contributions, and background, explaining odd expression patterns without inventing new cell states.
Results: in a mouse-brain dataset, the share of cells carrying both a neuronal and an insulating-cell marker dropped from 14.8% to below 0.01%.
Caveat: unusual signals aren't automatically errors, so corrections should still be checked against reference biology.

More from AGI Musings