AI helps identify new research directions in materials science
The rapidly growing number of scientific publications makes it increasingly difficult for researchers to keep track of emerging developments, even within their own field. In a recently published study in Nature Machine Intelligence, researchers from the Karlsruhe Institute of Technology (KIT), together with scientific partners including Prof. Christoph J. Brabec, demonstrate how artificial intelligence can support the identification of promising new research directions in materials science.
Materials science forms the basis of many key technologies, including photovoltaics, batteries, electronics, catalysis, and medical applications. However, the relevant knowledge is distributed across a vast and continuously expanding body of scientific literature. The study addresses this challenge by combining large language models with machine learning methods to systematically analyze scientific abstracts and detect connections between research concepts that may not yet have been explored.
The researchers analyzed approximately 221,000 materials science publications published between 1955 and 2022. Large language models were used to extract scientific concepts and chemical formulae from the abstracts. These concepts were then organized into a knowledge network, or concept graph, in which terms are represented as nodes and connections indicate how frequently concepts appear together in the literature. Machine learning models were subsequently trained to predict which previously unconnected concepts could become relevant combinations in the future.
The study shows that semantic information extracted by language models improves the prediction of emerging research topics. Importantly, the approach is not intended to replace scientific creativity, but to support it. By suggesting unexpected but potentially meaningful combinations of concepts, the method can help researchers look beyond established disciplinary boundaries and identify new opportunities for collaboration and discovery.
The practical relevance of the approach was evaluated through interviews with materials science experts, who assessed individualized AI-generated suggestions. Several of the proposed concept combinations were considered novel, interesting, or inspiring, demonstrating the potential of AI-assisted literature analysis as a tool for scientific ideation.
The publication highlights how artificial intelligence can contribute to future materials discovery by transforming the growing scientific literature into a structured source of new hypotheses and interdisciplinary research directions.
Original publication
Thomas Marwitz, Alexander Colsmann, Ben Breitung, Christoph Brabec, Christoph Kirchlechner, Eva Blasco, Gabriel Cadilha Marques, Horst Hahn, Michael Hirtz, Pavel A. Levkin, Yolita M. Eggeler, Tobias Schlöder, Pascal Friederich: Predicting new research directions in materials science using large language models and concept graphs. Nature Machine Intelligence 8, 535–544, 2026. DOI: 10.1038/s42256-026-01206-y