Scaling Biomedical Text-Mining: Transformers, GenAI, and Drug Discovery.

The exponential growth of biomedical literature has made manual curation and systematic knowledge extraction increasingly impractical, driving the need for robust and automated text mining. This chapter traces the evolution of biomedical text mining (BioTM) from rule-based and classical machine learning to deep learning, with a particular focus on transformer architectures and generative AI (GenAI). We examine how domain-adaptive pretraining, ontology-aware modeling, and long-context transformers improve foundational tasks (named entity recognition, normalization, and relation extraction) while reducing error propagation in multistage pipelines. We highlight the disruptive role of GenAI, including variational autoencoders, generative adversarial networks, diffusion models, and large language models, in hypothesis generation, molecular design, and knowledge graph construction, and summarize performance benchmarks and state-of-the-art applications in drug discovery and biomedical knowledge synthesis. We also surface open challenges: data coverage and bias, evaluation comparability and pretraining leakage, interpretability, computational cost, and ethical risks, including dual-use and privacy. Finally, we outline the regulatory context and emerging practices and emphasize the need for rigorous benchmarking, transparent models and data documentation, and multidisciplinary collaboration so that transformer-centered GenAI advances precision medicine while upholding scientific integrity and societal responsibility.