Understanding what life may look like on other planets has long been an aim of astrobiology. However, this aim has to date been hindered by a limited view of life constrained to Earth, which serves as the basis for environmental and organismal analogues to potential extraterrestrial counterparts. Many current efforts to detect life on other planets or beyond Earth rely on characteristics of life “as we know it”, constrained to the biochemistry of known Earth samples and analogues. Furthermore, estimates on the number of species as yet undiscovered even on Earth range widely, meaning that even the representative features of Earth-based life that are being used to guide life detection missions may be incomplete. Relying on the discovery of new physical examples of life to expand its definition significantly slows down the exploration of the potential state space. Challenging current approaches requires the use of novel techniques to expand the target ranges in the search for extraterrestrial life and biosignature detection. This project utilizes AI-based tools to explore the state space of chemicals of interest for biosignature and life detection studies.
Credit: Sunanda Sharma
Recent research has proposed non-photosynthetic pigments often found in extremophiles as potential surface reflectance biosignatures (Schweiterman et al. 2015), as well as outlined methods for the expansion of pigment families by structure and role, and rational design of pigments for optimized protective functionality (Schweitzer et al. 2009). We build upon these studies to propose a range of novel and unique pigment structures that are similar to but distinct from pigments found in Earth-based extremophiles, expanding the library of pigments under consideration as aspects of non-Earth based life.
Credit: Sunanda Sharma
The field of de novo molecular structure generation has been burgeoning as of late with the introduction of tools from deep learning and machine learning. These have found great promise in the area of drug discovery and designer molecule production, as they offer a path to intelligent discovery of organic molecules. In addition, the presence of ever-expanding databases such as PubChem, ChemBL, and others yield readily available training and testing datasets. Several deep learning architectures such as RNNs and GANs have been tested and applied, often using molecular representations such as SMILES. Recent innovation has been also conducted on these representations to provide more features relevant to structure-property connections in 3D space, rather than removing information for the sake of easy input.
We apply these techniques to the generation of non-Earth based synthetic molecules to test, validate, and quantify the degree of bias in life detection instrumentation.
Credit: Sunanda Sharma
Utilizing molecular modeling tools, approaches from AI-based drug discovery, and structure-property prediction techniques, we present a method to discover novel pigments and simulate their associated spectra relevant for identifying new model extremophiles as well as new potential surface biosignatures for life on other planets. Unlike physical experimental design methods, computational models offer a path for rapid iteration and library creation relevant for exploratory search and validation. In this way, we recontextualize and expand the scope of extremophile and biosignature research by drawing from seemingly disparate fields to accelerate progress and challenge current methods in key areas of astrobiology.
Credit: Sunanda Sharma