The first step in the analysis of multidimensional NMR spectra is based on the identification of spectral peaks, known as “peak picking”. The quality of this process has major consequences for all downstream analysis. Over the past several decades, different automated peak picking algorithms have been proposed to accomplish this task. Most of these algorithms are based on human-defined mathematical criteria to decide whether a data point is an actual peak or not. Although traditional peak picking methods have shown steady improvements, much of the NMR community still relies at least in part on manual peak picking by human experts considered as the gold standard. We recently demonstrated how DEEP Picker, a deep neural network (DNN) designed for spectral peak picking, is capable of deconvoluting highly crowded NMR spectra rivaling the facilities of human experts. Unlike traditional peak pickers, DEEP picker was trained, rather than explicitly coded, using a large synthetic database of NMR spectra of known composition with variable peak shapes and degrees of spectral overlap. Ideas behind the development of DEEP Picker will be discussed, such as how the training data was generated, how the neural network architecture was selected, why DEEP Picker can achieve superior accuracy, and how it can be extended to a wealth of different types of NMR spectra and their quantitative analysis.

Design of deep neural network for accurate deconvolution of complex NMR spectra

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.