DLQMA: A Deep Learning Framework for Qualitative and Quantitative NMR Analysis of Complex Hydrocarbon Mixtures

Nuclear magnetic resonance (NMR) is a key technique for structural and quantitative analysis of organic compounds, yet its application to complex mixtures is hindered by severe spectral overlap and the absence of efficient, standard-free quantification methods. DLQMA (Deep Learning for Qualitative and Quantitative Mixture Analysis) is introduced as a deep learning framework that simultaneously performs compound identification and concentration estimation directly from ¹H NMR spectra. Built on a pseudo-Siamese architecture with an added regression head, DLQMA enables end-to-end analysis without the need for manual spectral reconstruction or external standards. Performance was assessed using C8 hydrocarbon mixtures, a system characterized by severe spectral overlap and composed of multiple industrially relevant isomers with overlapping NMR signals. On this data set, DLQMA achieved 99.39% classification accuracy and a Pearson correlation of 0.98 for concentration predictions across 5,000 pairs of augmented validation spectra. The framework is further compatible with advanced NMR techniques such as 1D CSSF TOCSY, allowing automated interpretation of virtually separated subspectra. A standalone software implementation with an intuitive interface enables automated, high-throughput analysis of complex chemical mixtures. These advances have the potential to transform NMR spectroscopy into a fully automated tool for high-throughput analysis in chemical, petrochemical, and environmental applications.