A deep generative model enables automated structure elucidation of novel psychoactive substances

Skinnider, Michael A.; Wang, Fei; Pasin, Daniel; Greiner, Russell; Foster, Leonard J.; Dalsgaard, Petur Weihe; Wishart, David S.

doi:10.1038/s42256-021-00407-x

Cited by 37 publications

(43 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, Skinnider et al [ 58 ] applied a recurrent neural network-based generative model to attack a similar problem to that studied here, although they used a billion candidate structures and confined themselves to a test domain of some 2000 psychoactive substances. Although the approaches are not directly comparable, relatively few of their initial structures were accurate, although assessing them against the ‘structural priors’ in the billion molecules raised this to some 50%.…”

Section: Discussionmentioning

confidence: 99%

“…However, much of the problem of navigating chemical space in search of molecules that might match a given mass spectrum comes from the fact that chemical space is quasi-continuous but molecules are discrete [ 46 ]. As part of the revolution in deep learning [ 47 , 48 ], de novo generative methods have come to the fore (e.g., [ 46 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 ]). These admit the in silico creation of vectors in a high-dimensional ‘latent’ space (‘encoding’) and their translation from and into meaningful molecular entities (‘decoding’).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

MassGenie: A Transformer-Based Deep Learning Method for Identifying Small Molecules from Their Mass Spectra

et al. 2021

View full text Add to dashboard Cite

The ‘inverse problem’ of mass spectrometric molecular identification (‘given a mass spectrum, calculate/predict the 2D structure of the molecule whence it came’) is largely unsolved, and is especially acute in metabolomics where many small molecules remain unidentified. This is largely because the number of experimentally available electrospray mass spectra of small molecules is quite limited. However, the forward problem (‘calculate a small molecule’s likely fragmentation and hence at least some of its mass spectrum from its structure alone’) is much more tractable, because the strengths of different chemical bonds are roughly known. This kind of molecular identification problem may be cast as a language translation problem in which the source language is a list of high-resolution mass spectral peaks and the ‘translation’ a representation (for instance in SMILES) of the molecule. It is thus suitable for attack using the deep neural networks known as transformers. We here present MassGenie, a method that uses a transformer-based deep neural network, trained on ~6 million chemical structures with augmented SMILES encoding and their paired molecular fragments as generated in silico, explicitly including the protonated molecular ion. This architecture (containing some 400 million elements) is used to predict the structure of a molecule from the various fragments that may be expected to be observed when some of its bonds are broken. Despite being given essentially no detailed nor explicit rules about molecular fragmentation methods, isotope patterns, rearrangements, neutral losses, and the like, MassGenie learns the effective properties of the mass spectral fragment and valency space, and can generate candidate molecular structures that are very close or identical to those of the ‘true’ molecules. We also use VAE-Sim, a previously published variational autoencoder, to generate candidate molecules that are ‘similar’ to the top hit. In addition to using the ‘top hits’ directly, we can produce a rank order of these by ‘round-tripping’ candidate molecules and comparing them with the true molecules, where known. As a proof of principle, we confine ourselves to positive electrospray mass spectra from molecules with a molecular mass of 500Da or lower, including those in the last CASMI challenge (for which the results are known), getting 49/93 (53%) precisely correct. The transformer method, applied here for the first time to mass spectral interpretation, works extremely effectively both for mass spectra generated in silico and on experimentally obtained mass spectra from pure compounds. It seems to act as a Las Vegas algorithm, in that it either gives the correct answer or simply states that it cannot find one. The ability to create and to ‘learn’ millions of fragmentation patterns in silico, and therefrom generate candidate structures (that do not have to be in existing libraries) directly, thus opens up entirely the field of de novo small molecule structure prediction from experimental mass spectra.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

MassGenie: A Transformer-Based Deep Learning Method for Identifying Small Molecules from Their Mass Spectra

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Many of these studies have applied machine learning to analytical data to detect features of interest or as a classification tool; however, machine learning has been applied to the structures of NPS themselves. Skinnider et al ( 2021 ) reported the development of a deep generative model, termed DarkNPS, to predict novel structures using the known structures of NPS present on the HighResNPS database. The model was able to predict 176 of the 189 (93.1%) NPS-related compounds that were added to the database after the training set was finalised.…”

Section: Data Analysis and Machine Learningmentioning

confidence: 99%

Developments in high-resolution mass spectrometric analyses of new psychoactive substances

et al. 2022

Self Cite

View full text Add to dashboard Cite

The proliferation of new psychoactive substances (NPS) has necessitated the development and improvement of current practices for the detection and identification of known NPS and newly emerging derivatives. High-resolution mass spectrometry (HRMS) is quickly becoming the industry standard for these analyses due to its ability to be operated in data-independent acquisition (DIA) modes, allowing for the collection of large amounts of data and enabling retrospective data interrogation as new information becomes available. The increasing popularity of HRMS has also prompted the exploration of new ways to screen for NPS, including broad-spectrum wastewater analysis to identify usage trends in the community and metabolomic-based approaches to examine the effects of drugs of abuse on endogenous compounds. In this paper, the novel applications of HRMS techniques to the analysis of NPS is reviewed. In particular, the development of innovative data analysis and interpretation approaches is discussed, including the application of machine learning and molecular networking to toxicological analyses.

show abstract

“…Non-library matching methods generally involve two approaches. The first is a generative approach where synthetic GC-MS data is derived from theoretically predicted NPS molecular structures which then serves as references in a library for comparison [21]. Ji et al also reports a related method where molecular fingerprints are directly predicted from machine learning models and a molecular identity is predicted from the fingerprint combination [22].…”

Section: Introductionmentioning

confidence: 99%

Screening Unknown Novel Psychoactive Substances Using GC-MS Based Machine Learning

Wong

Tan

et al. 2022

Preprint

View full text Add to dashboard Cite

In recent years, there is a large increase in structural diversity of novel psychoactive substances (NPS), exacerbating drug abuse issues as these variants evade classical detection methods such as spectral library matching. Gas chromatography mass spectrometry (GC-MS) is commonly used to identify these NPS. To tackle this issue, machine learning models are developed to address the analytical challenge of identifying unknown NPS, using only GC-MS data. 891 GC-MS spectra are used to train and evaluate multiple supervised machine learning classifiers, namely artificial neural network (ANN), convolutional neural network (CNN) and balanced random forest (BRF). 7 classes, comprising 6 NPS classes (cathinone, cannabinoids, phenethylamine, piperazine, tryptamines and fentanyl) and other unrelated compounds can be effectively classified with a macro-F1 score ~ 0.9, averaged across all cross-validation folds. These results indicate that machine learning models are a promising complement as an effective NPS detection tool.

show abstract

A deep generative model enables automated structure elucidation of novel psychoactive substances

Cited by 37 publications

References 58 publications

MassGenie: A Transformer-Based Deep Learning Method for Identifying Small Molecules from Their Mass Spectra

MassGenie: A Transformer-Based Deep Learning Method for Identifying Small Molecules from Their Mass Spectra

Developments in high-resolution mass spectrometric analyses of new psychoactive substances

Screening Unknown Novel Psychoactive Substances Using GC-MS Based Machine Learning

Contact Info

Product

Resources

About