A Comparison of Different Approaches to Predicting Gas Chromatographic Retention Indices for Aromatic and Heterocyclic Compounds

Predicting retention indices is of great importance when improving the reliability of chromatography-mass spectrometry analysis. The main means of predicting was once linear equations, constructed for narrow classes of chemical compounds using molecular descriptors. Universal models based on neural networks trained with large databases, have in recent years acquired great importance. The problem of predicting retention indices by means of quantum chemistry has also been discussed in the literature. The aim of this work was to compare all these approaches using 45 aromatic compounds (mostly nitrogen-containing heterocycles) as an example. It is shown that the AIRI neural network is the most accurate and versatile model. Narrowly focused models based on linear regression and molecular descriptors can achieve almost the same (or even better accuracy) for a narrow class of compounds. The least resource-intensive means of quantum chemistry and molecular dynamics cannot reliably estimate the energy of intermolecular interactions, so predicting retention indices ab initio remains difficult for the foreseeable future.