Investors’ sneak peak: can this AI tool spot the science that will lead to patents?

Investors and technology-transfer offices are increasingly turning to artificial intelligence to identify promising scientific research that could lead to patents. A new machine-learning tool, known as the Translation Readiness Index (TRI), aims to predict which scientific papers are more likely to result in patents, months or even years before any formal patent filing or commercialization occurs. Developed by researchers at the data-analytics firm League of Scholars in Sydney, Australia, the tool analyzes the language used in a paper’s title and abstract to determine how closely it resembles previous publications that were later associated with patents. The TRI system was trained on a dataset comprising 20,610 scientific papers, of which 9,431 had been linked to patents. By comparing the vocabulary and phrasing of these papers, the algorithm identifies patterns commonly found in patentable research. The most effective model within the system achieved an accuracy rate of 78%, meaning it correctly ranked patent-linked papers higher than non-patent-linked ones when compared to similarly situated papers. This suggests that certain linguistic features, such as terms like “prototype,” “device,” and “design”, are frequently used in research that ultimately leads to patents. Paul McCarthy, a co-founder of League of Scholars and a co-author of the study, described the tool as a means of triaging research. He emphasized that while TRI provides a probabilistic assessment rather than a definitive prediction, it offers a valuable starting point for identifying potentially patentable work. The tool does not evaluate the actual experimental data or results contained within a paper, focusing instead on the language used to describe the findings. To validate the effectiveness of TRI, researchers examined the top 100 papers ranked by the tool among works authored by researchers at the University of Western Australia (UWA). These papers, published between 2019 and 2026, showed a significantly higher likelihood of featuring industry-affiliated co-authors or involving authors with prior patent experience. Specifically, 83 of the 100 papers had co-authors affiliated with industry, and 34 included at least one UWA-affiliated author who had previously filed a patent. Based on these findings, the team is currently collaborating with several universities to further refine and implement the tool. While McCarthy cautions against relying solely on TRI for investment decisions, he acknowledges that the tool could serve as a helpful filter for academic institutions and funding bodies seeking to prioritize high-potential research. Ben Miles, co-founder of Empirical Ventures, a London-based early-stage deep-tech investment firm, sees value in using TRI as an external signal for decision-making. He believes it could assist academics and funders in determining which projects warrant additional support before they reach a stage suitable for investor interest. Despite its promise, McCarthy notes that the ability to accurately measure commercial viability remains inherently uncertain. Patentable technologies do not always translate into profitable products, and thus any predictive tool must account for this limitation. Nevertheless, the growing demand for efficient methods to identify commercially relevant research continues to drive innovation in this space. Other institutions are also exploring similar approaches. For example, the technology team at Cornell University in Ithaca, New York, has developed a tool called Haystack to sift through the approximately 13,000 annual papers featuring Cornell-affiliated authors. With manual review proving impractical given the volume, tools like Haystack are becoming essential for identifying discoveries that could evolve into viable businesses.