A Bag of Tricks for Automated De Novo Design of Molecules with the Desired Properties: Application to EGFR Inhibitor Discovery
Source check: Needs source check. Not independent scientific confirmation.
Finding: A deep recurrent generator with reinforcement learning nominated EGFR compounds; enzyme assays confirmed 4 of 15 predicted actives, including two with nanomolar potency. Five predicted-inactive controls were inactive. This is a small preclinical proof of concept.
Technology: Generative recurrent neural network with reinforcement learning
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- A deep recurrent neural network generated SMILES molecules; reinforcement learning used an EGFR random-forest activity predictor with transfer learning, experience replay and reward shaping to guide generation and selection.
- Test and result
- The v1 preprint reports four confirmed inhibitors among 15 assayed predicted actives and five inactive predicted-inactive controls. Two predicted-active compounds were insoluble and not assayed. Two hits had nanomolar inhibition comparable to staurosporine; the selected-batch hit fraction does not establish a general success rate.
- Limitation / still unknown
- The result is a proof-of-concept assay on 15 predicted actives and five predicted-inactive controls. The preprint does not establish therapeutic efficacy or generalize its hit rate beyond this target and protocol.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://doi.org/10.26434/chemrxiv.14045072.v1
Needs source check · editorial details
The source has not been fully checked or contains conflicting information. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 18 Nov 2026. Review scope: Original ChemRxiv v1 PDF initial sections plus indexed primary assay/summary passages and official version metadata; complete PDF not independently re-read..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.