Structural characteristics and evolutionary trajectories of knowledge recombination in the field of AI-driven drug discovery

The rapid evolution of artificial intelligence (AI) and its profound integration with the pharmaceutical industry essentially constitute a process of knowledge recombination. Yet, empirical evidence systematically characterizing the structural and evolutionary dynamics of this phenomenon in AI-driven drug discovery remains scarce. Utilizing 494 granted invention patents from Chinese AI pharmaceutical firms (2011-2024), this study establishes a multi-method framework-integrating IPC co-occurrence networks, semantic similarity, LDA topic modelling, and longitudinal sliding-window analysis-to systematically characterize this cross-domain recombination. Structurally, the network exhibits a "sparse yet concentrated" topology. Bioinformatics (G16B) emerges as the critical bridging hub, facilitating distant recombination characterized by high combinatorial intensity but low cognitive similarity. Longitudinally, the network's evolution reveals a "temporal lag" in knowledge integration, marked by declining network density alongside a rising average degree. The technological dominance transitioned progressively from traditional pharmaceuticals (A61K) to computing (G06F), and ultimately to bioinformatics (G16B) and multidisciplinary integration. These findings offer an empirical lens into the micro-level dynamics of interdisciplinary knowledge convergence, providing preliminary insights that may inform capability development and innovation policymaking.