Artificial Intelligence and “Evolving” Databases: a Politecnico di Torino Study Proposes Thousands of New Materials for Energy Applications

We are pleased to share the official press release published by the Politecnico di Torino following the appearance of our peer-reviewed article in Energy and AI. This institutional announcement highlights the research conducted within the SMaLL – Multi-Scale Modeling Laboratory at DENERG and the creation of Energy-GNoME, an “evolving” AI-driven database for energy materials discovery.

Press Release — Politecnico di Torino

A team of researchers from the Politecnico di Torino — Paolo De Angelis, Giulio Barletta, Giovanni Trezza, Pietro Asinari, and Eliodoro Chiavazzo from the SMaLL – Multi-Scale Modeling Laboratory within the Department of Energy (DENERG) — has developed an innovative Artificial Intelligence (AI) protocol capable of selecting, among hundreds of thousands of previously unexplored materials, the most promising candidates for energy applications.

The study, published in the journal Energy and AI, introduces Energy-GNoME, the first “evolving” database that integrates Machine Learning algorithms with the valuable data from the GNoME (Graph Networks for Materials Exploration) project developed by Google DeepMind.

GNoME has recently made available to the scientific community an unprecedented resource: hundreds of thousands of previously unstudied and theoretically stable materials identified through generative AI techniques. However, these materials have not yet been “characterized” — that is, their potential technological applications have not been determined. This is precisely where Energy-GNoME comes into play: the method developed at the Politecnico makes it possible to identify, within the vast array of candidates proposed by GNoME, those with the highest potential for the energy sector, thus providing an essential bridge between the generation of new materials and their practical use.

The protocol adopts a two-step approach: first, a system of “artificial experts” that — through a majority vote — identifies compounds most likely to possess properties useful for energy applications; then, other dedicated models precisely estimate their key parameters. This method drastically reduces the number of candidates considered useful for a specific technological application, while at the same time proposing thousands of new solutions for energy conversion and storage.

“With Energy-GNoME, we wanted to demonstrate how Artificial Intelligence can be not just a tool for analysis, but a true accelerator of scientific discovery, capable of learning from human experience and growing through community contributions. At the same time, we aim to address a crucial challenge in generative AI: it is not enough to explore new possibilities blindly — it is necessary to direct this exploration toward useful objectives, because a crystal is only a chemical compound; it is its engineering function that makes it a material”, explains Paolo De Angelis, first author of the study.

“An important merit of the project lies precisely in the evolving nature of the database: through an open-source Python library and publicly available guidelines on GitHub, the scientific community will be able to contribute new experimental or theoretical data, feeding an iterative active-learning process. In this way, the platform is destined to evolve and constantly improve its predictive capability,” add Giulio Barletta and Giovanni Trezza.

“This approach represents a new frontier in materials modeling for energy applications: on one hand, it combines and leverages knowledge derived from experimental, theoretical, and machine-learning methods; on the other, it makes this synthesized knowledge available in an interoperable and accessible language, encouraging adoption and adaptation by different scientific communities,” adds Pietro Asinari.

“Our main contribution is twofold: on one side, making available to the scientific community a broad selection of new materials with strong potential for energy applications; on the other, developing a methodological protocol that can be easily extended to other domains beyond those explored in this study,” concludes Eliodoro Chiavazzo, coordinator of the research. “In this sense, Energy-GNoME is not just a database, but a true map to guide future experimental and computational studies, accelerating the exploration of advanced materials across multiple fields.”

Beyond its direct contribution to the energy domain, this work opens broader perspectives: the proposed protocol aims to become a methodological reference for the scientific community, offering a rapid and scalable path to explore new materials across diverse sectors — from advanced electronics to biomedicine, and from quantum technologies to emerging solutions for sustainability.