Prediction of transformative breakthroughs in biomedical research

The ability to predict scientific breakthroughs at scale would accelerate the pace of discovery and improve the efficiency of research investments. Recent advances in artificial intelligence, graph theory, and computing power have provided new ways to pursue this elusive goal. We have identified a common signature within co-citation networks that accurately predicts the occurrence of breakthroughs in medical research, on average more than 5 years in advance of the subsequent publication(s) that announced the discovery. A combination of features produces these diagnostic signals: a burst of papers exploring a novel scientific concept, an unusually high number of very influential papers in specialty journals, and low topical cohesion of the associated content. We analyzed two different periods separated by 20 years to show that the kinetics of breakthrough formation are conserved, suggesting that our approach can be used to predict which topics will produce future transformative discoveries.

Scientific breakthroughs are rare, as is contemporaneous recognition of their initial expression. Faster, more efficient identification of topics likely to produce future breakthroughs would speed scientific and technological progress. We introduce an AI/ML-detected signature in co-citation networks that recognizes such topics up to twelve years before the breakthrough itself occurs. Our findings illustrate how a better understanding of the scientific process may lead to greater scientific returns.