MolDiscovery: Learning Mass Spectrometry Fragmentation of Small Molecules

Abstract Identification of small molecules is a critical task in various areas of life science. Recent advances in mass spectrometry have enabled the collection of tandem mass spectra of small molecules from hundreds of thousands of environments. To identify which molecules are present in a sample, one can search mass spectra collected from the sample against millions of molecular structures in small molecule databases. This is a challenging task as currently it is not clear how small molecules are fragmented in mass spectrometry. The existing approaches use the domain knowledge from chemistry to predict fragmentation of molecules. However, these rule-based methods fail to explain many of the peaks in mass spectra of small molecules. Recently, spectral libraries with tens of thousands of labelled mass spectra of small molecules have emerged, paving the path for learning more accurate fragmentation models for mass spectral database search. We present molDiscovery, a mass spectral database search method that improves both efficiency and accuracy of small molecule identification by (i) utilizing an efficient algorithm to generate mass spectrometry fragmentations, and (ii) learning a probabilistic model to match small molecules with their mass spectra. We show our database search is an order of magnitude more efficient than the state-of-the-art methods, which enables searching against databases with millions of molecules. A search of over 8 million spectra from the Global Natural Product Social molecular networking infrastructure shows that our probabilistic model can correctly identify nearly six times more unique small molecules than previous methods. Moreover, by applying molDiscovery on microbial datasets with both mass spectral and genomics data we successfully discovered the novel biosynthetic gene clusters of three families of small molecules. Availability: The command-line version of molDiscovery and its online web service through the GNPS infrastructure are available at https://github.com/mohimanilab/molDiscovery.

Download Full-text

MolDiscovery: learning mass spectrometry fragmentation of small molecules

Nature Communications ◽

10.1038/s41467-021-23986-0 ◽

2021 ◽

Vol 12 (1) ◽

Author(s):

Liu Cao ◽

Mustafa Guler ◽

Azat Tagirdzhanov ◽

Yi-Yuan Lee ◽

Alexey Gurevich ◽

...

Keyword(s):

Mass Spectrometry ◽

Small Molecules ◽

Small Molecule ◽

Domain Knowledge ◽

Mass Spectra ◽

Molecular Structures ◽

Mass Spectral Database ◽

Mass Spectral ◽

Tandem Mass Spectra ◽

Small Molecule Databases

AbstractIdentification of small molecules is a critical task in various areas of life science. Recent advances in mass spectrometry have enabled the collection of tandem mass spectra of small molecules from hundreds of thousands of environments. To identify which molecules are present in a sample, one can search mass spectra collected from the sample against millions of molecular structures in small molecule databases. The existing approaches are based on chemistry domain knowledge, and they fail to explain many of the peaks in mass spectra of small molecules. Here, we present molDiscovery, a mass spectral database search method that improves both efficiency and accuracy of small molecule identification by learning a probabilistic model to match small molecules with their mass spectra. A search of over 8 million spectra from the Global Natural Product Social molecular networking infrastructure shows that molDiscovery correctly identify six times more unique small molecules than previous methods.

Download Full-text

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

10.20944/preprints202110.0355.v1 ◽

2021 ◽

Author(s):

Aditya Divyakant Shrivastava ◽

Neil Swainston ◽

Soumitra Samanta ◽

Ivayla Roberts ◽

Marina Wright Muelas ◽

...

Keyword(s):

Mass Spectrum ◽

Small Molecules ◽

Small Molecule ◽

Molecular Identification ◽

In Silico ◽

Mass Spectra ◽

Effective Properties ◽

Molecular Structures ◽

Mass Spectral ◽

Chemical Structures

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.

Download Full-text

MassGenie: a transformer-based deep learning method for identifying small molecules from their mass spectra

10.1101/2021.06.25.449969 ◽

2021 ◽

Author(s):

Aditya Divyakant Shrivastava ◽

Neil Swainston ◽

Soumitra Samanta ◽

Ivayla Roberts ◽

Marina Wright Muelas ◽

...

Keyword(s):

Mass Spectrum ◽

Small Molecules ◽

Small Molecule ◽

Molecular Identification ◽

In Silico ◽

Mass Spectra ◽

Effective Properties ◽

Molecular Structures ◽

Mass Spectral ◽

Chemical Structures

The ′inverse problem′ of mass spectrometric molecular identification (′given a mass spectrum, calculate 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. 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. 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.

Download Full-text

Computational methods for small molecule identification

it - Information Technology ◽

10.1515/itit-2019-0033 ◽

2019 ◽

Vol 61 (5-6) ◽

pp. 285-292

Author(s):

Kai Dührkop

Keyword(s):

Mass Spectrometry ◽

Tandem Mass Spectrometry ◽

Statistical Model ◽

Small Molecules ◽

Computational Methods ◽

Small Molecule ◽

Mass Spectra ◽

Tandem Mass ◽

Molecular Fingerprint ◽

Fragmentation Tree

Abstract Identification of small molecules remains a central question in analytical chemistry, in particular for natural product research, metabolomics, environmental research, and biomarker discovery. Mass spectrometry is the predominant technique for high-throughput analysis of small molecules. But it reveals only information about the mass of molecules and, by using tandem mass spectrometry, about the mass of molecular fragments. Automated interpretation of mass spectra is often limited to searching in spectral libraries, such that we can only dereplicate molecules for which we have already recorded reference mass spectra. In my thesis “Computational methods for small molecule identification” we developed SIRIUS, a tool for the structural elucidation of small molecules with tandem mass spectrometry. The method first computes a hypothetical fragmentation tree using combinatorial optimization. By using a Bayesian statistical model, we can learn parameters and hyperparameters of the underlying scoring directly from data. We demonstrate that the statistical model, which was fitted on a small dataset, generalizes well across many different datasets and mass spectrometry instruments. In a second step the fragmentation tree is used to predict a molecular fingerprint using kernel support vector machines. The predicted fingerprint can be searched in a structure database to identify the molecular structure. We demonstrate that our machine learning model outperforms all other methods for this task, including its predecessor FingerID. SIRIUS is available as commandline tool and as user interface. The molecular fingerprint prediction is implemented as web service and receives over one million requests per month.

Download Full-text

Study on Mass Spectral Database Obtained by Liquid Chromatography/Mass Spectrometry for Retrieval System of Chemicals

BUNSEKI KAGAKU ◽

10.2116/bunsekikagaku.54.211 ◽

2005 ◽

Vol 54 (3) ◽

pp. 211-219 ◽

Cited By ~ 1

Author(s):

Michiko UEBORI ◽

Yoshiaki ISHII ◽

Atsuko HASEGAWA ◽

Yasuko YOSHIDA ◽

Shigeru SUZUKI ◽

...

Keyword(s):

Mass Spectrometry ◽

Liquid Chromatography ◽

Retrieval System ◽

Liquid Chromatography Mass Spectrometry ◽

Mass Spectral Database ◽

Mass Spectral ◽

Chromatography Mass Spectrometry ◽

Spectral Database

Download Full-text

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

Biomolecules ◽

10.3390/biom11121793 ◽

2021 ◽

Vol 11 (12) ◽

pp. 1793

Author(s):

Aditya Divyakant Shrivastava ◽

Neil Swainston ◽

Soumitra Samanta ◽

Ivayla Roberts ◽

Marina Wright Muelas ◽

...

Keyword(s):

Mass Spectrum ◽

Small Molecules ◽

Molecular Identification ◽

In Silico ◽

Mass Spectra ◽

De Novo ◽

Effective Properties ◽

Molecular Structures ◽

Mass Spectral ◽

Chemical Structures

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.

Download Full-text

Differentiation of the Isomeric o-(m- and p-) Nitro-(Chloro- and Bromo-)Benzyl-2,4-(and 2,1-) Disubstituted 2-Thiocytosinium Halides by Electron Impact Ionisation and Fast Atom Bombardment Mass Spectrometry

European Journal of Mass Spectrometry ◽

10.1255/ejms.996 ◽

2009 ◽

Vol 15 (4) ◽

pp. 497-506 ◽

Cited By ~ 2

Author(s):

Tomasz Pospieszny ◽

Elżbieta Wyrzykiewicz

Keyword(s):

Mass Spectrometry ◽

Mass Spectra ◽

Fast Atom Bombardment ◽

Collision Induced Dissociation ◽

Mass Measurements ◽

Fragmentation Pathways ◽

Mass Spectral ◽

Fragment Ions ◽

Accurate Mass Measurements ◽

Impact Ionisation

Electron ionisation (EI) and fast atom bombardment (FAB) mass spectral fragmentations of nine 2,4-(and 2,1-) disubstituted o-( m- and p-)nitro-(chloro- and bromo-)-2-thiocytosinium halides are investigated. Fragmentation pathways, whose elucidation is assisted by accurate mass measurements and metastable transitions [EI-mass spectrometry (MS)], as well as FAB/collision-induced dissociation (CID) mass spectra measurements are discussed. The correlations between the abundances of the (C11H10N4SO2)+1–3; (C11H10N3SCl)+4–6 and (C11H10N3SBr)+7–9 ions and the selected fragment ions (EI-MS), as well as (C18H16N5SO4)+1–3; (C18H16N3SCl2)+4–6 and (C18H16N3SBr2) + 7–9 ions and the selected ions (C7H6NO2)+1–3; (C7H6Cl)+ 4–6; (C7H6Br)+ 7–9 (FAB-MS) are discussed. The data obtained can be used for distinguishing isomers.

Download Full-text