The prediction of dipole moments is essential to obtain spectral information and optical forces for any system. Recent experimental work has shown that optical forces mediated by e.g. plasmonic or cavity environments can be efficiently used to influence the chemical reactivity in SN2 reaction (see https://doi.org/10.1038/s41467-022-35363-6 and references therein). In this project, we utilize a machine learning approach based on neuro-evolution potentials (NEP), a type of neural network-based machine learning model, to learn and predict the dipole moments of PTA structures in ionic solution. In combination with nuclear forces obtained via the same NEP approach, it provides a way to estimate the reactivity in various ambient conditions at a massively discounted cost. First steps have already been taken by obtaining course grained density-functional theory (DFT) data to obtain first NEP potentials. The obtained IR spectra are consistent with DFT calculations at low energies with increasing deviations at larger energies, suggesting the need for an improved model at further away from the local minimum. Initial tests to estimate the reactivity are successful but indicate the need for more refined models. In conclusion, the project requires improved NEP models and an active learning approach (iterative evaluation and retraining of the NEP models), which requires more computational resources.