Bayesian Models and Biomarker detection in astrobiology

Since the confirmation of the first exoplanet in 1992, exoplanetary science has advanced significantly, with over 5600 confirmed planets in the NASA database. A central goal of the field is to understand habitability and identify biomarkers for life detection. However, the science of biosignature currently lacks a universally accepted theoretical framework to make rigorous life detection claims. My thesis focuses on the application of the Bayesian framework to biosignature detection and highlights both its strengths and limitations. The Bayesian framework offers a flexible and quantitative tool for evaluating life-detection claims, but currently his reliability seems to be hindered by the uncertainties on prior probability of life P(life), which remains poorly constrained due to the absence of a robust theoretical framework. Furthermore the application of Bayesian model shows that in the scenario where P(molecule|life) << P(molecule|nolife), a high posterior probability of life can only be achieved if the prior probability is already very high, which is a rarely realistic assumption. In such cases repeated observations lead to a steep decline in posterior probability rather than an increase. Only eliminating false positives or selecting biomarkers with no false positives, combined with large statistical samples, can lead to high-confidence detection over time. The results obtained from sensitivity analysis using first order Sobol indices highlights the necessity of detecting multiple biosignatures to reduce the dependency on prior assumptions, making high-confidence life-detection claims more attainable. Finally, as a case study, in the last chapter I analyzed the alleged detection of Phosphine in Venus, highlighting the challenges in life-detection and how the lack of application of a Bayesian framework makes it difficult to rigorously establish confidence in life detection claims.