Microsoft Research: discarding singleton proteins in PLM training is likely wrong
KevinKaichuang · x · 2026-08-18
A Microsoft Research / Brown preprint argues the common practice of discarding metagenomic "singletons" in PLM pretraining is probably wrong. Singletons make up 43% of the 3.34B-sequence GigaRef atlas; many show high homology and higher Pfam domain density per length, representing the long tail of functional diversity most valuable for model training.
More from Research
- MIT PhD Sharut Gupta joins DeepMind; Isola hails her representational alignment defense talk — phillip_isola · 2026-10-16
- 36 primitives, just 10 parameters each: a remarkably precise T-pose body fit — letandrewcook · 2026-10-16
- Interp researcher pushes back: if you can't simulate a neuron, you haven't explained it — nabla_theta · 2026-10-16
- MLP-Splatting (ECCV 2026) code released: MLPs as light-field primitives instead of Gaussians — AjdDavison · 2026-10-16
- Four-page solution to subcubic APSP builds intuition for Anthropic-AI-aided breakthrough — gautamcgoel · 2026-10-16
- ATAS: Zhejiang-led team builds world-state memory framework for consistent agentic video generation — CSProfKGD · 2026-10-16