A new way to predict how cells react to genetic and chemical changes

The scBIG++ framework predicts how all protein-coding genes change when cells face interventions.

Researchers built scBIG++, a framework to predict single-cell transcriptional responses. This tool predicts how cells react to genetic and chemical interventions.

The method uses Gene-Relation Clustering to organize genes into common windows. Dynamic SoftSub learns expression-dependent submodule memberships within these windows. Set-GCAE exchanges information about gene groups across control cells.

The system uses a latent flow initialized by cell representations. A pooled summary and the perturbation representation condition the response generation. This process decodes states into expression profiles for all assayed protein-coding genes.

Experiments used six genetic and chemical settings. These settings spanned 6,5396,539--19,45119,451 genes. The framework showed competitive recovery of perturbation effects and full-gene expression reconstruction.

The tool works for unseen targets and combinations. It also works for cross-cell-line transfer and held-out compounds.

Key facts

  • scBIG++ predicts responses for all assayed protein-coding genes.
  • The framework was tested on 6,5396,539--19,45119,451 genes.
  • The method works with unseen targets, combinations, and held-out compounds.

All sections

Edition