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

Abstract

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.

Ensemble Spectral Prediction (ESP) Model for Metabolite Annotation

“…CSI:FingerID utilizes fragmentation trees that best explain the spectra (31,32) to improve mapping candidates to their corresponding fingerprints (9,33). In addition to learning to rank, we expect auxiliary information in various forms, such as biochemical (34) data or data augmentation (10) to improve annotation.…”

“…L = −cos(ŷ, y) [10] where PRED(•) is neural network that predicts spectra from molecular representation z, and cos(•, •) is the cosine similarity between the predicted spectra, ŷ, and the query spectra, y. For all PRED(•) functions, we applied a two-layer MLP augmented with bidirectional prediction mode (17), which increases the prediction accuracy on the larger fragments that arise due to neutral losses.…”

“…CSI:FingerID (9) predicts properties molecular fingerprints for a set of candidate molecules, and then ranks the query spectra against the predictions. MassGenie (10) casts the spectrum-to-molecule problem as a translation problem from binned mass spectral peaks to a SMILES string using the deep-learning transformer model (11). MassGenie then uses a variational autoencoder to generate de novo candidate molecules that are 'close' in the chemical space.…”

Bi-modal Variational Autoencoders for Metabolite Identification Using Tandem Mass Spectrometry

“…Future work will seek to optimize and tune the model itself. Current deep learning methods that predict information from the mass spectra use binned spectra and process them with multilayer perceptrons 20,31 , word2vec algorithm inspired by natural language processing 32 or a transformer architecture 57 . A binned spectrum has an obvious drawback of reducing resolution of the original spectrum, losing sensitivity provided by the latest generation of mass spectrometers.…”

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

“…We believe that it represents the first use of transformers in molecular identification from (mass) spectra. A preprint (dated 26 June 2021) has been lodged [ 84 ].…”