ACCELERATING SCIENTIFIC DISCOVERY WITH GENERATIVE KNOWLEDGE EXTRACTION, GRAPH-BASED REPRESENTATION, AND MULTIMODAL INTELLIGENT GRAPH REASONING

Leveraging generative AI, the study transforms a dataset of 1,000 biological materials papers into a comprehensive ontological knowledge graph. Structural analysis (node degrees, communities and their connectivities, clustering coefficients, and betweenness centrality) reveals a highly connected, inherently scale-free network useful for downstream reasoning. Using large language model embeddings to compute deep node representations and combinatorial node similarity ranking, the authors develop a path sampling strategy that links previously dissimilar concepts. Examples include structural parallels between biological materials and Beethoven’s 9th Symphony via isomorphic mapping and a proposed hierarchical mycelium-based composite inspired by Kandinsky’s Composition VII, integrating chaos/order balance, adjustable porosity, strength, and patterned chemical functionalization. The method uncovers cross-domain isomorphisms across science, technology, and art, revealing nuanced, context-dependent heterarchical interplay. By transcending disciplinary boundaries and modalities (graphs, images, text, numerical data), the graph-based generative AI approach achieves high novelty and technical detail, revealing hidden connections for querying, gap identification, materials design, and behavior prediction.

Markus J. Buehler