400-model study finds single-cell foundation models lack clear data scaling laws
sebkrier · x · 2026-11-14
A new study pretrained 400 models on a corpus of 22.2M cells and ran 6,400 evaluations, finding current single-cell foundation models plateau at much smaller training sizes without language-model-like data scaling laws. Collecting more of the same single-cell data isn't the answer—coverage of diverse biological states, conditions, perturbations and endpoints matters more.
More from Research
- Paper proposes 1500-word briefs plus machine-verifiable appendices to fix AI-era peer review — WalterReade · 2026-11-14
- EnvOpt: Optimizing Agent Environments Beats Bigger Models on Cost and Pass Rate — AlexGDimakis · 2026-11-14
- Researcher uses a coding agent to fetch latest Marin scaling-law fit parameters — RylanSchaeffer · 2026-11-14
- World Mechanics launches inaugural fellowship for multimodal interpretability research in 2027 — soniajoseph_ · 2026-11-14
- humans& open-sources a hardware-native NVFP4 4-bit RL training recipe — gharik · 2026-11-14
- NVIDIA BioNeMo's NV-deCYPher takes 3rd on both OpenADMET Blind Challenge tracks — AllThingsApx · 2026-11-14