AI trained on cell experiments guides mRNA discovery: 4x liver expression over benchmark lipid in mice
bravo_abad · x · 2026-10-13
Zhang et al. trained an XGBoost classifier on protein-expression measurements from cultured human cells treated with 520 different lipids, then used it to prioritize candidates from a virtual library of 10,920 designs for mRNA delivery.
The selected candidates were independently tested in mice: at a matched dose, the lead formulation produced roughly fourfold higher reporter-protein expression in the liver than the benchmark lipid SM102, despite similar liver accumulation—suggesting the gain came from intracellular delivery rather than targeting.
Notably, the model predicted activity only in cultured cells, yet the improvement transferred to living animals, showing the value of ML-guided experiment selection from in vitro data.
More from Research
- EXL details GNN fraud detection system with 27-33% detection gains — AI Engineer · 2026-10-13
- RTNav: open-source modular asynchronous agent for real-time zero-shot object navigation — rsasaki0109 · 2026-10-13
- AI-CFD benchmark culture: a mesh-free solver takes a cow to Mach 2 — burhop · 2026-10-13
- The Implicit Curriculum Hypothesis wins COLM outstanding paper award, powered by NSF public compute — IanArawjo · 2026-10-13
- Computational biology shifts as AI builds 'better predictors' in under a day — jmschreiber91 · 2026-10-13
- Model weights are only 1 of 4 VRAM allocations: why loaded LLMs still OOM — blaizedsouza · 2026-10-13