<p>In this paper, we describe a non-invasive method of diamond identification that exploits intrinsic defects of natural crystals embedded into the crystal lattice during their formation. The inevitable lattice defects manifest themselves through the spectral dependence of light absorption in the mid-IR region of the optical spectrum. By collecting multiple spectral data taken from different parts of the diamond sample, followed by the data analysis using methods of deep machine learning, we are able to identify spectral markers associated with otherwise unique lattice defect pattern. The core of the proposed method is a neural network trained on a number of samples belonging to the same class. The class could be multiple measurements of the same diamond or gemstones of the same geographical origin. We demonstrate that the trained neural network is able to identify previously untested samples and to correctly predict their origin. Results of our study suggest that the underlying diamond property allowing its identification is ever-present native defects in the lattice. The three-dimensional pattern of defects, set by temperature, pressure, and their gradients in both time and space during crystal formation, appears to hold information about diamond identity, including its origin. In many aspects, it resembles genomics in living cells. Our results suggest that each diamond mine is associated with a unique set of spectral markers that can be used for cost-effective and potentially very fast non-invasive identification of diamond origin.</p>