Conditional Molecular Generation Net Enables Automated Structure Elucidation Based on ¹³C NMR Spectra and Prior Knowledge

Abstract

Structure elucidation of unknown compounds based on nuclear magnetic resonance (NMR) remains a challenging problem in both synthetic organic and natural product chemistry. Library matching has been an efficient method to assist structure elucidation. However, it is limited by the coverage of libraries. In addition, prior knowledge such as molecular fragments is neglected. To solve the problem, we propose a conditional molecular generation net (CMGNet) to allow input of multiple sources of information. CMGNet not only uses 13C NMR spectrum data as input but molecular formulas and fragments of molecules are also employed as input conditions. Our model applies large-scale pretraining for molecular understanding and fine-tuning on two NMR spectral data sets of different granularity levels to accommodate structure elucidation tasks. CMGNet generates structures based on 13C NMR data, molecular formula, and fragment information, with a recovery rate of 94.17% in the top 10 recommendations. In addition, the generative model performed well in the generation of various classes of compounds and in the structural revision task. CMGNet has a deep understanding of molecular connectivities from 13C NMR, molecular formula, and fragments, paving the way for a new paradigm of deep learning-assisted inverse problem-solving.

Impact of noise on inverse design: the case of NMR spectra matching

“…Note that the knowledge of two functional groups only led to marginal improvements. However, fragmentation could be more beneficial for larger compounds than present in QM9, 1 as reported by Yao et al 58 Using both 13 C and 1 H shifts on the reduced search space only lead to marginal improvements of 0.5% over the results of the full search space.…”

Advances in the Application of Artificial Intelligence-Based Spectral Data Interpretation: A Perspective

“…Spectral data can be regarded as continuous sequences, and the application of RNNs in natural language processing (NLP) provides useful guidance for processing spectral data; ab initio generation especially provides another perspective for interpreting spectral data. 18 Convolutional Neural Networks (CNNs). Convolutional neural networks (CNNs) were first developed for computer vision tasks and are now widely used in facial recognition, image detection, and other fields.…”

“…Spectral data can be regarded as continuous sequences, and the application of RNNs in natural language processing (NLP) provides useful guidance for processing spectral data; ab initio generation especially provides another perspective for interpreting spectral data …”

“…In this study, a QM calculation-based method was used for predicting the spectrum of the generated SMILES molecule and comparing it with the target spectrum, and the comparison result was returned to update the prefix used for the RNN training. Recently, Yao et al proposed a BART-based conditional generation model called “conditional molecular generation network” (CMGNet), which enables multiconditional input to be used and generates structures based on 13 C NMR data, molecular formulas, and fragment information (Figure ). Because the model accepted molecular formulas and fragment information as output conditions, the model generated structures based on 13 C NMR data and the recovery rate reached 94.17% for the top 10 candidate compounds.…”

Cross-Modal Retrieval Between 13C NMR Spectra and Structures Based on Focused Libraries

“…To overcome the shortcomings of an oversize library, we introduced CMGNet, a generative model based on bidirectional and autoregressive transformers, to provide focused libraries for each identification test (Figure ). , CReSS achieved a 9.17% Top-10 accuracy based on a random reference structure library, whereas CFLS achieved a Top-10 accuracy of 54.03% in the test involving the analysis of 6,471 13 C NMR spectra. In addition, CFLS also exhibited the ability to sort the correct structure in a more prominent rank.…”