Comparative biosignatures with systemic retrievals

TL;DR

The paper tackles the challenge of distinguishing biogenic from abiotic biosignatures in exoplanet atmospheres by introducing a comparative, system-wide framework of systemic retrievals that defines a local abiotic baseline within planetary systems. It develops a Bayesian formalism in which abiotic and biotic forward models are compared using leave-one-out predictive densities (elpd_LOO) and an evidence-based Delta E_L threshold to identify potential biological anomalies. The approach integrates abundance-space and raw data-based inferences within a hierarchical Bayesian model, enabling marginalisation over latent, shared planetary and stellar parameters to robustly test biosignatures across multiple planets. It further extends to habsignatures and habiosignatures, outlines practical diagnostics for model comparison, and discusses the construction of super-systems to enhance statistical power for system-level biosignature detection.

Abstract

The discovery of inhabited exoplanets hinges on identifying biosignature gases. JWST can reveal biosignature gases, though current discoveries have yet to evidence life. The central challenge is attribution: how can we confidently identify biogenic sources while ruling out, or deeming unlikely, abiotic explanations? Attribution is particularly difficult for individual planets, especially given the stochastic abiotic processes that can set atmospheric conditions. To address this, we propose a comparative multi-planet approach centred on systemic retrievals: the analysis of multiple planets within a system to empirically define the `abiotic baseline'. This baseline, constructed from obligate uninhabited planets, serves as a local reference point. Systemic retrievals enable marginalisation over inaccessible, latent, shared abiotic parameters within planet evolution models. This is possible because planets within a system are linked by their birth in the same natal disk, have been irradiated by the same evolving star, and have a linked dynamical history. Observations aligning with the abiotic baseline, where the locally-informed abiotic planet evolution models demonstrate high out-of-sample predictive accuracy, are likely non-biological. Potentially biological anomalies are identified as statistical outliers from the abiotic baseline using Bayesian leave-one-out cross-validation. A comparative biosignature is thus defined: an anomaly where a biotic planetary evolution model provides a superior fit than its abiotic counterpart. Where both abiotic and biotic models yield poor predictive accuracy, the anomaly is flagged as an ``unknown unknown"; a signature of either unconstrained abiotic processes, or life as we don't yet know it.

Figures (9)

• Figure 1: Earth's atmospheric composition as an example of comparative biosignatures and habsignatures. Earth, situated within the habitable zone (green annulus sector), emerges as an outlier. Compared to the abiotic baseline established by Venus and Mars, Earth exhibits significantly lower CO2 levels thienen2007ingeology and considerably higher O2 levels (indicated below each planet). All of these gases have plausible abiotic origins, but the abiotic baseline informs our expectation and identifies Earth as anomalous.
• Figure 2: Schematic of the Geochemical Planet Evolution Model for Systemic Retrievals. This workflow illustrates how observations of a planetary system are used to constrain a forward model of planetary evolution. The model takes directly observed stellar and systemic parameters $\theta^*$, shared planetary parameters $\theta^s$ and planetary parameters $\theta^p$, such as mass (M$_p$), radius (R$_p$), and semi-major axis (a$_p$), and observed stellar and systemic parameters ($\theta^*$) as inputs. A key feature of this systemic approach is its ability to simultaneously model and marginalise over latent parameters --- crucial properties that are hidden from direct observation. These include latent planetary properties (e.g., cloud properties, impact flux) and, most importantly, latent system-shared parameters $\theta^*$ and $\theta^s$ that are inaccessible when analysing planets in isolation. The geochemical model links all these parameters, both observed and latent, to produce simulated observables for each planet in the system. The final output can be either simulated spectra (S$^p$) or the inferred atmospheric abundances ($x^p$) with their associated uncertainties, which are then compared to actual observations.
• Figure 3: Procedural construction of the abiotic baseline. Step 1: Atmospheric observations are collated for the planetary system. While the inference framework operates on high-dimensional data (e.g., spectra or abundance profiles), these are visualised here as projected scalar signal strengths. Step 2: The Hierarchical Bayesian Model marginalises over shared latent parameters to constrain system-wide abiotic trends. Step 3: The abiotic baseline is iteratively defined by excluding high-influence outliers (Pareto $k > 0.7$). Step 4: These anomalies are evaluated against biotic models; those yielding $\Delta \mathcal{E}_L > 4$ are classified as comparative biosignatures.
• Figure 4: Illustration of comparative biosignatures using abiotic baselines. Planetary systems 1 and 2 are represented by their respective abiotic baselines, B$_1$ (purple line) and B$_2$ (green line). Candidate biosignature (or signal strength) observed for planets within each system are shown: purple circles for System 1 and green squares for System 2. The two filled symbols are planets within the habitable zones of their respective systems. The x-axis represents a systemic parameter (e.g., planetary mass, orbital distance); the y-axis represents observational data, which for illustration purposes are simplified to a single data point per planet. The top right panel highlights how even if two planets occupy a similar region in parameter space (i.e., show signs of biosignatures), their contextualisation within their respective planetary system baselines differs. For the case of the HZ planet from B$_2$ (filled green square), the planet's biosignature signal strength falls within agreement of the system's baseline. Conversely, for the HZ planet from B$_1$ (purple filled circle) it deviates significantly from its system's abiotic baseline, so presents a possible biotic anomaly. Such anomalies are comparative biosignatures, potentially indicative of the planets in the system having been modified by life.
• Figure 5: The Bayesian hierarchical model linking latent parameters to observables. Stellar and systemic parameters ($\theta^*$) influence shared planetary parameters ($\theta^s$), which in turn determine planet-specific parameters ($\theta^p$) for each planet $p$ of $N_p$. These parameters collectively shape the observed stellar and planetary context, $C_{\text{obs}}$ (as referenced in Section \ref{['sec:bayes']}). The presence or absence of life on a planet (life$^p$), along with all $\theta$ parameters, influence the abundances of atmospheric species ($x^p$), which in turn determine the planet’s observed spectral or photometric data (D$^p$). The models $\mathcal{M}$ and $\mathcal{M}_L$ represent abiotic and biotic planet evolution forward models that simulate the observations.