Performance of Organ-Focused Versus Generalized Single-Cell Foundation Models

Determine whether kidney-specific single-cell foundation models pretrained on organ-focused datasets outperform generalized multi-organ single-cell foundation models in terms of accuracy and interpretability within single-cell analysis.

Background

Foundation models such as scGPT and Geneformer have demonstrated strong performance across diverse single-cell tasks by pretraining on large, multi-organ datasets. However, the extent to which organ-specific biological structure can be better captured by specialized, organ-focused models has not been conclusively established.

The kidney’s complex cellular architecture and dynamic microenvironments present challenges for model generalization, batch correction, and cross-modality integration. This motivates the question of whether a kidney-focused foundation model can surpass generalized models in accuracy and interpretability for kidney single-cell analysis.

References

Large foundation models have revolutionized single-cell analysis, yet no kidney-specific model currently exists, and it remains unclear whether organ-focused models can outperform generalized models.

— Nephrobase Cell+: Multimodal Single-Cell Foundation Model for Decoding Kidney Biology  (2509.26223 - Li et al., 30 Sep 2025) in Abstract

A plausible reading is that exposure to many scanners and protocols forces contrast to be encoded in shared terms, whereas a narrower brain-only corpus does not, although with two models per group this remains a hypothesis.

— Dataset-Specific Encoding of Image Content Underlies Cross-Dataset Failure in Brain MRI Foundation Models  (2610.13027 - Luu et al., 9 Oct 2026) in Section 6, Discussion