Recommendations for molecules of interest in the Taurus Molecular Cloud-1

This dataset contains 1510 molecules of interest to the chemical inventory of TMC-1, recommended based on grounds of chemical similarity. These molecules were identified as part of work currently under review:

Lee et al., "Machine Learning of Interstellar Chemical Inventories" (2021)

Structures were generated from their SMILES strings using OpenBabel and rdkit, and geometry optimization carried out using the geomeTRIC program:

Wang, L.‑P.; Song, C.C. (2016), J. Chem, Phys. 144, 214108. http://dx.doi.org/10.1063/1.4952956

Electronic structure calculations were performed using psi4, with both geometry optimization and dipole moments calculated at at the 𝜔B97X‑D/6‑31+G(d) level of theory. Equilibrium dipole moments
and rotational constants are reported in unsigned debye and MHz respectively; for the latter, we provide effective scaled parameters as well that empirically correct for vibration‑rotation interactions. Please refer to “Lee, K. L. K. and McCarthy, M. 2020, J Phys Chem A, 5, 898” for information regarding their uncertainties. For molecules where SCF/geometry optimizations failed to converge, we provide their dipole moments based on the molecular mechanics structures. These molecules will be indicated by “Is DFT optimized?: False”.

The predicted column densities and uncertainties are given with a simple Gaussian Process with rational quadratic and white noise kernels. Simply put, the predicted column densities of unseen molecules are given as functions of distance in chemical space that decays naturally to zero for infinite distance from other data points. The reader is encouraged to look at the distances between recommendations and TMC‑1 molecules to develop an intuition for how the predicted column density behaves roughly with distance, and interpret them with the uncertainties accordingly: as a guide but not to rule out molecules specifically. Molecules with particularly large uncertainties are
likely to be impactful in constraining the chemistry of the source, even if we provide just an upper limit.

Finally, there is no real ordering of which the molecules are given. This is quasi‑random, although there are pockets of similar molecules based on how similar the TMC‑1 molecules are between searches.

The included tarball contains the OpenBabel generated XYZ coordinates for each recommended species. This github repository contains the notebooks and code used to generate this data.