scRM introduces a single-cell foundation model trained on sequencing reads

The scRM preprint presents a reference-free model trained directly on single-cell RNA-seq reads and says raw-read pretraining is competitive at a fraction of the computational cost. Its public repository contains model/training/evaluation code and states MIT licensing; the README limits the described pipeline to 10x Chromium single-cell polyA assays, excluding probe-based Flex and spatial assays. The preprint has not been peer reviewed, and this capture does not verify public pretrained-weight availability or reproduce results.