A graph-attentive GAN for rare-cell-aware single-cell RNA-seq data generation.

GARAGE uses graph attention to generate realistic synthetic single-cell RNA-seq data, improving rare cell representation and downstream analysis accuracy.

• Why it matters: Accurate analysis of scRNA-seq data is hindered by high-dimensionality, small sample sizes, and rare cell types, limiting feature selection and clustering effectiveness.
• What they did: The method integrates a graph attention network with a GAN, injecting attention-weighted real cell embeddings into the generator to focus on under-sampled populations, across multiple benchmarks.
• The result: GARAGE enhances feature selection and clustering, preserves rare-cell structures, accelerates training, and reduces mode dropping, enabling more reliable biological insights from scRNA-seq data.