A neural-network model for finding correlations in astrochemical space

The detection of new interstellar molecules is a central objective of astrochemistry, and arguably the most transformative development in the field over the past decade has been the dramatic increase in both the number and diversity of detected species. Using state-of-the-art natural language processing models, we developed a machine-learning pipeline, in line with recent approaches reported in the literature, to establish relationships between molecular structures, encoded as chemical strings, and molecular column densities in the prototypical dark cloud TMC-1. The workflow generates astrochemically informed embeddings that can be employed in subsequent supervised learning tasks. In this work, these representations are used within a neural-network framework to perform column-density regression on both the original molecular inventory and out-of-distribution species. The resulting model achieves mean squared error values of 0.17 and 0.58 and mean absolute error values of 0.29 and 0.59, all expressed in logarithmic units of the column density, together with coefficients of determination of 0.88 and 0.51 for the train+validation and validation datasets, respectively. The latent vectors learned by the model represent a rich mathematical object that captures not only the underlying chemical relations but also the potential interstellar detectability of the molecule. Using these vectors in molecules obtained from systematic permutations of the TMC-1 inventory allowed us to identify several promising candidates for future detection in this cloud, together with a large number of apolar species. Furthermore, by training the model on datasets reflecting different historical epochs of molecular detections in TMC-1, we show that recent molecular detections have substantially reshaped our understanding of interstellar chemistry in a transformative, rather than incremental, way. Beyond the accurate prediction of column-density trends for both in-distribution and out-of-distribution molecules, the latent-space representations produced by the model provide a stepping stone for developing new forms of data-driven astrochemical intuition and exploration.