PanFoMa Achieves Superior Pan-Cancer Single-Cell Modeling
Researchers introduce PanFoMa, a lightweight hybrid neural network combining Transformers and linear-time state-space models, in a Dec 2, 2025 preprint. They also release PanFoMaBench, a curated pan-cancer single-cell benchmark of over 3.5 million cells across 33 cancer subtypes. PanFoMa outperforms prior models by +4.0% on the pan-cancer benchmark and shows gains in cell type annotation (+7.4%), batch integration (+4.0%), and multi-omics integration (+3.1%).
Key Points
- 1Introduces PanFoMa, a hybrid Transformer and state-space model for efficient single-cell transcriptome representation learning.
- 2Demonstrates consistent performance gains: +4.0% pan-cancer benchmark, +7.4% cell-type annotation, +4.0% batch integration.
- 3Enables scalable analyses via PanFoMaBench with 3.5 million cells across 33 cancer subtypes.
Scoring Rationale
Novel hybrid architecture and large benchmark deliver strong empirical gains, limited by preprint status and domain-specific scope.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
