SB-CFG replaces the standard unconditional branch in scRNA-seq diffusion models with a deliberately under-informative sparse reference. This approach avoids retaining gene-specific structures from the marginal distribution, improving the effectiveness of conditional generation for single-cell RNA sequencing.
HOW THIS AFFECTS YOU
●
researcherThis provides a new guidance strategy for diffusion models where the marginal distribution is poorly defined.
●
healthThis may lead to more accurate synthetic single-cell genomic datasets for biological modeling.