Virtual Gene Knockout
The authors of a preprint compared eight methods that attempt to predict the consequences of gene knockout based on single-cell RNA data. In a test on K562 leukemia cells, the direction predicted by the linear version of CellOracle matched the experiment in 18 out of 44 "transcription factor — gene" pairs. The data on single-cell RNA shows which genes are usually active together, allowing researchers to build a hypothesis about the regulatory network and choose a gene for the next experiment.
The actual gene knockout answers a different question: how will the work of each specific gene change after intervention. The authors tested four transcription factors — proteins that control the work of other genes — and 11 glycolysis genes, the first stage of a cell obtaining energy from glucose. In Perturb-seq, researchers use CRISPRi to suppress the work of a selected gene and then read the RNA of individual cells. The average activity of these 11 genes decreased in all four factors.
The measurements were then compared to the signs of coefficients in the linear version of CellOracle. The sign of the coefficient describes the relationship between genes in the RNA data, and CRISPRi shows their response to intervention. The directions matched in 18 out of 44 cases, as reported in the preprint of August 19.
🔗 Read original →
August 24, 2026 10
