Breakthrough in exoplanet science: 44 new candidates for Earth-like worlds

A research team from the University of Bern and the National Center of Competence in Research PlanetS has achieved a significant milestone in the search for habitable planets. As announced on April 9, 2025, the team developed a machine learning model capable of pinpointing planetary systems likely to contain Earth-like exoplanets with remarkable accuracy. This breakthrough not only advances the hunt for potentially habitable worlds but also marks a promising step toward the discovery of extraterrestrial life.

The AI model was developed under the guidance of Dr. Jeanne Davoult as part of her doctoral research at the University of Bern, with support from Prof. Dr. Yann Alibert and Romain Eltschinger from the Center for Space and Habitability (CSH). It was trained using synthetic data generated by the renowned "Bern Model of Planet Formation and Evolution," which simulates the physical processes underlying the formation of planetary systems. The outcome is striking: with a reported 99% accuracy, the model successfully identified systems that are highly likely to contain at least one Earth-like planet.

Real-world application to planetary systems

After training, the model was applied to real observational data and identified 44 planetary systems that could potentially host previously unknown Earth-like planets. These findings are particularly significant for upcoming space missions like ESA’s PLATO and the proposed LIFE project, both of which aim to detect and characterize Earth-like worlds.

PLATO (PLAnetary Transits and Oscillations of stars), set to launch in 2026, will use the transit method and asteroseismology to detect potentially habitable exoplanets, with a focus on those orbiting sun-like stars. The most promising candidates identified by PLATO will form the basis for future missions such as LIFE (Large Interferometer For Exoplanets), which aims to analyze the atmospheres of distant planets using infrared spectroscopy and nulling interferometry to search for biosignatures like water or methane. The new machine learning model could play a key role in pre-selecting the most promising targets, thereby enhancing the efficiency and success rate of these missions.