scGPT Extracts Compact Hematopoietic Algorithm With High Performance
Researchers report on Mar 10, 2026 the discovery and extraction of a compact hematopoietic algorithm from the single-cell foundation model scGPT using mechanistic interpretability. They introduce a three-stage operator-export method producing a standalone, zero-shot transferable algorithm that outperforms scVI, Palantir, DPT and other baselines on 88-split donor-holdout benchmarks (CD4/CD8 AUROC 0.867; mono/macro AUROC 0.951) while running 34.5× faster with ~1000× fewer trainable parameters.
Scoring Rationale
High novelty and strong practical efficiency, but limited by a single preprint source lacking peer-reviewed validation across diverse cohorts.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
