From fragments to drugs: How AI and in silico methods accelerate fragment-to-lead optimization

Fragment-based drug design (FBDD) has become a key approach for mapping chemical space by starting with small, low-affinity fragments and building them into potent leads. Although these fragments bind weakly, strategies like fragment merging and linking can convert them into high-affinity molecules. More recently, Artificial Intelligence (AI) and Machine Learning (ML) have sped up this workflow by powering structure-guided optimization and generative design of new compounds. This review presents the state of FBDD: from biophysical screening and rapid structure elucidation to fragment growing, merging, and linking that elevate affinity while preserving ligand efficiency and drug-like properties. In particular, this review focuses on the integration of AI/ML approaches within these workflows. We compare experimental and computational strategies, summarize representative case studies, and assess how AI/ML now supports hit triage, compound priorization, and efficient exploration of chemical space. By emphasizing the role of AI-driven methods within established FBDD pipelines, we aim to provide a perspective on how these approaches complement traditional strategies. Limitations, common artifacts, and validation practices are discussed to clarify what reliably works and where open challenges remain in FBDD.