Nephrobase Cell+: Multimodal Single-Cell Foundation Model for Decoding Kidney Biology.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Large foundation models have advanced single-cell analysis, but a kidney-specific model is lacking.
• Integrating kidney single-cell and spatial omics data is challenging due to cellular complexity and batch effects.
• Existing frameworks struggle with cross-modality variation and cross-species generalization in kidney data.

Purpose of the Study:

• To develop the first kidney-focused large foundation model.
• To evaluate if organ-specific models outperform generalized models for kidney data.
• To improve the integration and analysis of multi-modal kidney omics data.

Main Methods:

• Developed Nephrobase Cell+, a kidney-specific foundation model using a transformer-based encoder-decoder architecture.
• Pretrained the model on approximately 100 billion tokens from 39.5 million single-cell and single-nucleus profiles across four mammalian species.
• Incorporated gene-token cross-attention and a mixture-of-experts module for scalable representation learning.

Main Results:

• Nephrobase Cell+ established a new benchmark for kidney single-cell analysis, producing biologically coherent embeddings.
• Achieved superior cluster concordance and batch-mixing scores, effectively removing batch effects while preserving cell-type structure.
• Demonstrated >90% zero-shot annotation accuracy for major kidney lineages across species, outperforming existing models.

Conclusions:

• Nephrobase Cell+ provides a unified, high-fidelity representation of kidney biology.
• The model is robust, cross-species transferable, and unmatched by current single-cell foundation models.
• Offers a powerful resource for advancing kidney genomics and disease research.