AI for Astrophysics & Planetary
Science

An Interactive Exploration of the Cosmos

Flag of Israel Assaf Shaked · Flag of Brazil Juliana Merege · Flag of India Yankit Kukreja

Flag of Switzerland Colin Gross · Flag of Spain Michael Cebral · Flag of Puerto Rico Raul Mariani

Flag of Canada Vanessa Van Decker

Scroll to Explore

A long time ago, in a galaxy far far away the search for habitable worlds began...

Today this search is powered by mountains of data from missions like Kepler.

Yet, most Kepler planets are too faint to study their Planetary Habitability Index (PHI) characteristics directly, and ask, “Could life live here?”

However, JWST excels at atmospheric spectroscopy. By linking JWST’s insights with Kepler’s vast catalog, AI could unlock advanced, low-cost habitability research.

The challenge? This data belongs to individual teams, but cross-mission links could extend JWST’s reach to many more worlds.

This project marks the next step in exoplanet research, moving from AI-based detection to large-scale PHI analysis.

AI for Astrophysics & Planetary Science

We have found a candidate!
But we need more information...
Let's get closer to see what we find...

The Kepler Space Telescope

Launched in 2009 and decommissioned in 2018, NASA´s Kepler mission revolutionized exoplanet discovery. The Kepler database holds brightness data for about 150,000 stars; by analysing this data, you and scientists can spot tiny dips, called "transits" in the stars' light curves, which suggest the presence of exoplanets orbiting those stars.

An artist's impression of the Kepler Space Telescope.

Earth Similarity Index (ESI)

Provides a quantitative 0–1 score measuring an exoplanet’s physical resemblance to Earth in terms of mass and radius. ESI ≥ 0.8 is typically considered “Earth-like.” Although it is not a measure of actual habitability, ESI is useful for filtering large datasets such as the Kepler catalog. Below is an interactive chart showing the concept of transit.

The James Webb Telescope

As the successor to Hubble, the James Webb Space Telescope is the most powerful and complex space observatory ever built. Its massive, gold-coated mirror and advanced infrared instruments allow it to peer deeper into the cosmos than ever before, capturing the faint light from the very first stars and galaxies, and analyzing the atmospheres of distant exoplanets for signs of life.

An artist's impression of the James Webb Space Telescope.

Objectives:

Objective 1:

Increase the rate of exoplanet candidate validation and PHI characterization by developing AI-driven models.

Objective 2:

Establish a cross-mission collaboration framework between space telescopes for unified habitability analysis.

Methods

How do we build a robust, data-driven pipeline for finding habitable worlds?

Stage 1: Broad Screening

A deterministic ESI score is calculated from Kepler's catalog data, while a CNN analyzes light curves to assess signal plausibility.

Kepler Data
(Free & Open Source)
↓

Pass/Fail Gate

A candidate passes Stage 1 only if ESI ≥ 0.8 AND its Light Curve Score ≥ 0.5.

Dual-Criteria Filter

~95% of candidates filtered

Stage 2: Targeted Refinement

AI models, pre-trained on JWST data, analyze Kepler parameters to infer atmospheric properties and calculate a PHI Likelihood score for final ranking.

AI Trained on
JWST Data
↓

Final Shortlist

A prioritized list of high-potential targets for in-depth biosignature investigation.

High-Confidence Candidates

Results and Discussion

Conclusions

Future Mission Integration
  • This AI-driven framework is designed for scalability and can readily integrate data from upcoming missions like the Nancy Grace Roman Space Telescope, PLATO, and the Habitable Worlds Observatory. A multi-mission, multi-wavelength approach will enhance the reliability of our candidate selection by cross-validating detections and refining planetary parameters.

AI-Driven Habitability Framework
  • Our two-stage pipeline illustrates a crucial strategy for modern exoplanet science: leveraging the strengths of different observatories. Stage 1 uses survey telescopes for broad, efficient screening with physics-based metrics (ESI) and signal validation (LC CNN). Stage 2 applies resource-intensive assets like JWST for targeted, in-depth analysis of the most promising candidates using sophisticated spectral and likelihood models.

Prioritized Target Selection
  • The primary outcome is a dynamically updated, statistically robust shortlist of high-potential targets. By acknowledging the reality that most Kepler targets lack JWST spectra, our pipeline correctly prioritizes candidates that either have existing data or show exceptional promise based on MLP likelihood, ensuring that precious observatory time is allocated with maximum efficiency.

Reproducibility and Scalability
  • This framework emphasizes reproducibility by embedding all model weights, preprocessing parameters, and decision thresholds directly within the application. This ensures that the analysis is transparent and repeatable, establishing a consistent methodology for comparing candidates across different missions and accelerating the collective scientific search for habitable worlds.

References

  1. Alderson, L., Sing, D.K., Doyon, R., Spake, J.J., Evans, T.M., Lagage, P.O., et al. (2025) A JWST transit of a Jupiter analog. Nature, 631, pp.95–100. DOI: 10.1038/s41586-025-09876-2.
  2. Ansdell, M., Ioannou, Y., Osborn, H.P., Sasdelli, M., Smith, J.C., Jenkins, J.M., Raissi, C. & Angerhausen, D. (2018) ‘Scientific domain knowledge improves exoplanet transit classification with deep learning’. Available at: https://arxiv.org/abs/1810.13434
  3. Chopra, A. and Lineweaver, C.H. (2016) The case for a Gaian bottleneck: The biology of habitability. Astrobiology, 16(1), pp.7–22. Mary Ann Liebert, New Rochelle, USA. DOI: 10.1089/ast.2015.1387.
  4. Malik, A., Moster, B.P. & Obermeier, C. (2021) ‘Exoplanet detection using machine learning’. Available at: https://arxiv.org/abs/2011.14135
  5. Rao, S., Mahabal, A., Rao, N. & Raghavendra, C. (2021) ‘Nigraha: Machine-learning based pipeline to identify and evaluate planet candidates from TESS.’ Monthly Notices of the Royal Astronomical Society, 502(2), pp.2845–2858. DOI: 10.1093/mnras/stab203. Available at: https://arxiv.org/abs/2101.09734
  6. Shallue, C.J., Thompson, S.E. & Christiansen, J.L. (2021) Single transit detection in Kepler with machine learning and onboard spacecraft diagnostics. Publications of the Astronomical Society of the Pacific, 133(1031), 064501.
  7. Schulze-Makuch, D., Méndez, A., Fairén, A.G., von Paris, P., Turse, C., Boyer, G., Davila, A.F., Resendes de Sousa António, M., Catling, D. & Irwin, L.N. (2011) A two-tiered approach to assessing the habitability of exoplanets. Astrobiology, 11(10), pp.1041–1052. DOI: 10.1089/ast.2010.0592.
  8. Valizadegan, H., Martinho, M.J.S., Wilkens, L.S., Jenkins, J.M., Smith, J.C., Caldwell, D.A., Twicken, J.D., Gerum, P.C.L., Walia, N., Hausknecht, K., Lubin, N.Y., Bryson, S.T. & Oza, N.C. (2022) ExoMiner: A highly accurate and explainable deep learning classifier that validates 301 new exoplanets. Astrophysical Journal, 926(2), 120.
  9. PHL @ UPR Arecibo - Habitable Worlds Catalog [WWW Document], n.d. URL https://phl.upr.edu/hwc (accessed 11.10.25).
  10. Schulze-Makuch, D., Méndez, A., Fairén, A.G., von Paris, P., Turse, C., Boyer, G., Davila, A.F., António, M.R. de S., Catling, D., Irwin, L.N., 2011. A Two-Tiered Approach to Assessing the Habitability of Exoplanets. Astrobiology 11, 1041–1052. https://doi.org/10.1089/ast.2010.0592.

Extra Resources

Full Abstract

The search for habitable worlds beyond our Solar System has evolved rapidly, driven by the legacy of the Kepler Space Telescope in 2009. Even after retirement in 2018, Kepler’s dataset continues to inspire methodologies in exoplanet detection, classification, and validation through artificial intelligence (AI) and cross-mission collaboration.

Yet, most Kepler planets remain too faint and distant for direct habitability characterization, constrained by signal-to-noise limits and orbital geometry. This can be addressed through the combined use of James Webb Space Telescope (JWST) data and AI-driven analysis.

Research has begun exploring correlations between Kepler’s photometric data and atmospheric properties by comparing light-curve features with simulated refraction patterns. This line of inquiry demonstrates that Kepler’s archive can serve as the foundation for habitability inference when enhanced by JWST observations.

A challenge is that the intellectual property (IP) rights of JWST biosignature images typically remain with individual principal investigators, making large-scale AI training datasets difficult to assemble. Nonetheless, shared access would yield an invaluable resource for exoplanet science.

AI capabilities open the door to integrating datasets from different space observatories, allowing models to bridge missions and accelerate the detection of potentially habitable exoplanets. The integration of scientific domain knowledge into deep-learning classifiers proved that hybrid astrophysical–neural models outperform purely data-driven systems.

This project proposes an interactive platform that unites recent advances in AI and astrophysical modelling while recommending integration of spectral and physical parameters with a two step process utilizing the Earth Similarity Index (ESI) and Planet Habitability Index (PHI).

Very recently, the first cross-mission analysis between Kepler and JWST focused on Kepler-167e Jupiter-like exoplanet, demonstrating that Kepler’s archival photometric data can be directly integrated with infrared observations.

This project aims to reflect a logical progression: from AI-powered detection to detailed habitability study, establishing a framework for comparing distant worlds with our own.

Click for a cosmic experience!