Quantitative mixture analysis is a universal objective across all scientific arenas. Most reactions, consumer products, and formulations are complex mixtures of a number of known and unknown components. Accessing concentration information in an efficient and robust framework is critical in discovery and commercialization pipelines. The advent of modern programming accessibility and computational power has given experimentalists unprecedented access to sophisticated algorithms and mathematical machinery which can drastically improve spectral resolution. NMR spectroscopy is particularly well-poised for broad adoption of these machine leaning tools considering the plethora of spectral filtration and decoupling techniques available to work in concert with digital deconvolution techniques. This talk will focus on an unconventional paradigm – how NMR experiments can be purposely tailored for machine learning. Non-Fourier techniques such as Hadamard spectroscopy and non-uniform sampling (NUS) can be judiciously employed to partially deconvolute spectra. Orchestration of these bespoke excitation profiles and modern signal processing algorithms can produce spectra with exquisite resolution ready for efficient downstream analysis.
