Machine learning struggles with drug library predictions, study finds
Researchers discovered that AI models trained on DNA-encoded chemical libraries fail when tested against novel compounds, limiting their use in drug discovery. The findings expose a critical flaw in current computational approaches and offer a roadmap for building more reliable prediction tools that could accelerate pharmaceutical development.
Originaltitel: Toward generalizable predictive models for DNA-encoded libraries
• Machine learning applied to DEL data accelerates hit identification and scaffold exploration. • DEL-trained ML models often struggle to generalize to novel chemical scaffolds, as shown in AURKA case studies due to domain shift. • Methodological best practices and benchmarking with open DEL datasets improve model robustness and guide future generalizable DEL-ML development. DNA-encoded libraries (DELs) combined with machine learning (ML) offer a powerful paradigm for hit identification. However, sequencing-derived enrichment data are inherently noisy and biased, often resulting in models that overfit to specific chemical libraries. In this review, we critically evaluate the capabilities and limitations of DEL-ML, illustrating key challenges using Aurora Kinase A (AURKA) DEL affinity selection data. We demonstrate that standard ML models often struggle to generalize to unseen chemical space because of the specific structural constraints of combinatorial libraries. Furthermore, we discuss the necessity of rigorous denoising strategies and evaluate approaches, such as domain adaptation, to mitigate these limitations, offering a roadmap for building robust models capable of exploring diverse chemical space.