Comparing AI and Established Methods to Mine Medicinal Chemistry Patents for Discovery Insights

Presentation for BioIT World, Boston, May 2026 Medicinal chemistry patents represent an underused resource for drug discovery. Between 2000 and 2024, the per-year medicinal chemistry-centric WO filings rose from 2,636 to 9,521 and include bioactivity data for millions of structure–activity relationships (SAR). Established tools such as OPSIN and DECIMER can convert chemical names and images to structures; PubChem now hosts 48 million patent-extracted compounds including 30 million from SureChEMBL and 22 million from PATENTSCOPE that enable text and chemistry searches. In addition, semi-automated SAR curation from by BindingDB (PMID39574417) has submitted over 1,2 million patent-extracted binding measurements to PubChem BioAssay. Recent advances in AI have the potential to extend the transformation of this data into actionable knowledge. Paradoxically, the combination of new and established workflows are making patent data FAIR-er than paywalled literature, unlocking new opportunities to accelerate discovery.