Lunai Bioworks reports 0.88 AUROC in structure-based AChE inhibitor screening model using NIH Tox21 data
LNAI•Broader safety-review program
The effort builds on an April 2026 commercial defense program pairing mechanism-of-action prediction with in vivo zebrafish assays for neuroactive compound classification.
Model-agnostic screening approach
Work targets model-agnostic screening that runs independently of the LLM or molecule generator, to triage candidates for expert review or testing.
AChE inhibition model results on NIH Tox21 data
AChE inhibition modeling on NIH Tox21 data tested 9,667 compounds with scaffold-split sets; held-out results showed AUROC 0.88.
At one operating point, precision reached 33% at 70% recall, implying 5.1-fold enrichment versus 6.4% AChE-active prevalence.
BioSymetrics case study on structure-based screening
Lunai Bioworks released a BioSymetrics case study on structure-based screening to flag chemical outputs from generative AI for safety review.
