RNA language model trained on sequence alone uncovers new human riboswitches
IgorCarron · x · 2026-09-26
Harvard Medical School's Silvi Rouskin lab released an updated preprint on bioRxiv: an RNA language model, never explicitly taught thermodynamics, trained purely on sequences, learns RNA structural logic on its own.
- It recovers alternative riboswitch structures and critical tertiary loop-loop contacts without supervision
- It also surfaces new putative human riboswitches, with more results teased
- The paper's core contribution is revealing the structural logic of Internal Ribosome Entry Sites (IRES); authors span Harvard Medical School, Mass General Brigham, and Hampshire Therapeutics
A notable AI-for-Science result: sequence-only training suffices to infer RNA folding and functional elements, no structural labels required.
More from Research
- Stanford and NVIDIA's CLM embeds decisions instead of tokens, 13x faster but accuracy drops at scale — Prompt Engineering · 2026-09-26
- Researchers skip ICLR over review chaos — and once a journal collapsed before deciding — Aaroth · 2026-09-26
- Salesforce research: extracting agent memory just-in-time beats summarizing, +16 success rate — dair_ai · 2026-09-26
- MIT CSAIL's JAZ: minimalist agent with one 'invoke' primitive beats MemGPT at half the cost — omarsar0 · 2026-09-26
- Michael Levin team preprints a pipeline to compensate ion channel mutations with drugs — drmichaellevin · 2026-09-26
- Fact-Checking the NeurIPS-Pangram Story: v3.3.2 Flagged 14.9% of Human-Written Reviews as AI — tughanbulut · 2026-09-26