# scBIG++

Single-cell perturbation prediction has largely been evaluated on highly variable genes, leaving coordinated expression changes in excluded genes unavailable for downstream analysis. scBIG++ extends prediction to all protein-coding genes retained after preprocessing, assigns gene contributions dynamically from each cell's expression within shared gene windows, and exchanges information about the same windows across control cells. On the Norman additive setting its full-gene response correlation is 0.8521 versus 0.8160 for scBIG, with MSE falling from 0.0017 to 0.0015.

Single-cell perturbation prediction has largely been evaluated on highly variable genes, leaving coordinated expression changes in excluded genes unavailable for downstream analysis. scBIG++ extends prediction to all protein-coding genes retained after preprocessing, assigns gene contributions dynamically from each cell's expression within shared gene windows, and exchanges information about the same windows across control cells. On the Norman additive setting its full-gene response correlation is 0.8521 versus 0.8160 for scBIG, with MSE falling from 0.0017 to 0.0015.