scRM: A single-cell foundation model trained on scRNA-seq reads

scRM is the first single-cell foundation model trained directly on scRNA-seq sequencing reads, showing raw-data pretraining is feasible and competitive at a fraction of the computational cost.

• Why it matters: Count-based approaches leave information in raw sequencing signals untapped, and the field lacks a read-level foundation model to test whether pretraining on raw data can help.
• What they did: The authors train scRM on scRNA-seq sequencing reads and compare performance and compute cost against count-based modeling, emphasizing feasibility and efficiency.
• The result: The results suggest scaling scRM and trying alternative architectures could reveal biological information not accessible to count models and enable dataset-specific pretraining.

AI summary of the abstract · may contain errors · not medical advice · read the paper.