Highly Selective Novel Heme Oxygenase-1-Targeting Molecules Discovered by DNA-Encoded Library-Machine Learning Model beyond the DEL Chemical Space

DNA-encoded library (DEL) technology has proven to be a powerful method for discovering novel inhibitors for diverse targets. Particularly when combined with machine learning (ML), the DEL-ML workflow expands the chemical space and enhances cost-effectiveness, offering new opportunities to find desired hit molecules. Heme oxygenase-1 (HO-1), primarily a heme-degrading enzyme, has been identified as a potential therapeutic target in diseases such as cancer and neurodegenerative disorders. Despite years of study, the HO-1 inhibitor toolbox remains limited. Here, we report the discovery of five series of novel scaffold HO-1 inhibitors using a DEL-ML workflow that emphasizes the model’s uncertainty quantification and its domain of applicability. The DEL-ML model demonstrated a strong ability to extrapolate to novel chemical spaces by identifying new structures. Approximately 33% of the predicted molecules, validated by biophysical assays, had a binding affinity of K¬D < 15 µM, with the strongest affinity being 141 nM. Fourteen tested molecules showed over 100-fold selectivity towards HO-1 over Heme oxygenase-2 (HO-2). These molecules are also structurally novel compared to the reported HO-1 inhibitors. Further, binding mode simulations via docking provided insights into the possible selectivity rationale of some selective series.