First look at Nanome Agents
Directing agents at a target to discover novel molecules or peptides worth ordering
The week before this session we pointed our agents at ALK2 R206H, the kinase variant behind fibrodysplasia ossificans progressiva, and let them run.
Together, we used a journal article to directly pull SMILES strings, convert 2D molecular drawings into SMILES, extract SAR data from tables in the paper, and enrich the dataset with ChEMBL data. Our Agents prepped receptors from the RCSB, found pockets with Pocketeer, and verified structure prediction results from Boltz2, Protenix, and OpenFold3 by comparing them to the original crystal structure.
- Thursday, September 3, 2026
- 9:00 AM PT / 12:00 PM ET · 60 min
- Online, no headset needed
The Nanome Agents then identified molecular cores and enumerated new analogs to backfill the limited number of starting structures, docked the expanded dataset with GNINA, and used the resulting docking scores as rewards in a de novo REINVENT4 workflow to generate novel candidates. We repeated nine cycles using only newly generated compounds. At the end of each cycle, the top 20 compounds were co-folded with Boltz2 and OpenFold3.
Join us as we recap the process: docking score variation, peptide binder designs, and the ADMET gates used to triage potential hits before running molecular dynamics, MM/GBSA, and MM/PBSA calculations.
We also ran a complete de novo binder design campaign using our proprietary agent-guided algorithm, and we can't wait to share the data from it!