SIMPLE: Sparse Interaction Model over Peaks of moLEcules for fast, interpretable metabolite identification from tandem mass spectra

Abstract

MotivationRecent success in metabolite identification from tandem mass spectra has been led by machine learning, which has two stages: mapping mass spectra to molecular fingerprint vectors and then retrieving candidate molecules from the database. In the first stage, i.e. fingerprint prediction, spectrum peaks are features and considering their interactions would be reasonable for more accurate identification of unknown metabolites. Existing approaches of fingerprint prediction are based on only individual peaks in the spectra, without explicitly considering the peak interactions. Also the current cutting-edge method is based on kernels, which are computationally heavy and difficult to interpret.ResultsWe propose two learning models that allow to incorporate peak interactions for fingerprint prediction. First, we extend the state-of-the-art kernel learning method by developing kernels for peak interactions to combine with kernels for peaks through multiple kernel learning (MKL). Second, we formulate a sparse interaction model for metabolite peaks, which we call SIMPLE, which is computationally light and interpretable for fingerprint prediction. The formulation of SIMPLE is convex and guarantees global optimization, for which we develop an alternating direction method of multipliers (ADMM) algorithm. Experiments using the MassBank dataset show that both models achieved comparative prediction accuracy with the current top-performance kernel method. Furthermore SIMPLE clearly revealed individual peaks and peak interactions which contribute to enhancing the performance of fingerprint prediction.Availability and implementationThe code will be accessed through http://mamitsukalab.org/tools/SIMPLE/.

ADAPTIVE: leArning DAta-dePendenT, concIse molecular VEctors for fast, accurate metabolite identification from tandem mass spectra

“…A drawback of ADAPTIVE would be interpretability, because structural information is implicitly encoded in compact vectors in ADAPTIVE and cannot be made explicit easily. In metabolite identification, it would be desirable to connect the set of peaks to the corresponding substructures/chemical properties of metabolites (Nguyen et al , 2018b). Developing a model with such interpretability would be interesting future work.…”

“…Thus, we can say that kernel-based supervised learning, particularly complex kernels, have a computation issue, regardless of high performance in prediction. On the other hand, a sparse learning model, namely SIMPLE (Nguyen et al , 2018b), considers a simpler function than kernels for fingerprint, while interactions of peaks in spectra can be incorporated into learning models explicitly. SIMPLE achieved a comparable performance against kernel-based learning, reducing the computational cost drastically.…”

“…Recent advances in metabolite identification have been led by the machine learning category (e.g. Brouard et al , 2016; Dührkop et al , 2015; Nguyen et al , 2018b). This category can be further divided into two key groups: supervised learning for substructure prediction and unsupervised learning for substructure annotation.…”

“…The prediction can be divided into two steps: (i) fingerprint prediction: predicting fingerprints of a given test spectrum with supervised learning; (ii) candidate retrieval: retrieving chemical compound (from database) which is closest to the predicted fingerprints (Nguyen et al , 2018b).…”

Probabilistic Framework for Integration of Mass Spectrum and Retention Time Information in Small Molecule Identification

“…In recent years numerous powerful approaches (Nguyen et al, 2018a;Schymanski et al, 2017) for annotating MS 2 spectra with a predicted molecular structure have been developed (Ruttkies et al, 2016(Ruttkies et al, , 2019Dührkop et al, 2015;Brouard et al, 2016;Allen et al, 2014;Nguyen et al, 2018bNguyen et al, , 2019Dührkop et al, 2019). Typically, these methods output a ranked list of molecular structure candidates, that can be shown to human experts, or further post-processed, e.g.…”