<p>Summary of the datasets utilized here.</p>

These factors hinder robust feature (gene) selection and cell clustering and limit the realism of samples generated by existing simulators. We introduce <i>GARAGE</i>, a <b>G</b>raph-<b>A</b>ttentive <b>RA</b>re-cell aware single-cell data <b>GE</b>neration that augments the generator’s input with a small, attention-weighted <i>‘leakage’</i> of real cells in addition to prior noise.

Specifically, we build a <i>k</i>-nearest-neighbour cell graph and use a graph attention network (GAT) to prioritize nodes that likely represent under-sampled (rare) subpopulations; these high-attention cell embeddings are injected into the generator input to steer synthesis toward biologically plausible regions of the data manifold while respecting cell-type proportions. This attention-guided leakage accelerates training, reduces mode dropping, and yields realistic synthetic cells that preserve rare-cell structure.