Comparison of various K-nearest neighbor voting schemes with the self-training interpretive and retrieval system for identifying molecular substructures from mass spectral data

Lowry, Stephen R.; Isenhour, T. L.; Justice, Joseph B.; McLafferty, Fred W.; Dayringer, Henry E.; Venkataraghavan, R.

doi:10.1021/ac50020a022

Cited by 24 publications

(14 citation statements)

References 15 publications

(37 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…At present, three strategies for structural classification exist: a) Cluster compounds based on spectral similarity, then propagate compound class annotations from database search in a semiautomated manner [14][15][16] b) Search for the query compound in a spectral library 17,18 or a structure database 19,20 ; consider the top k hits for assigning compound classes. c) Use machine learning methods to directly predict compound classes from the MS/MS spectrum 19,21 .…”

Section: Introductionmentioning

confidence: 99%

Classes for the masses: Systematic classification of unknowns using fragmentation spectra

Dührkop

Nothias

Fleischauer

et al. 2020

Preprint

View full text Add to dashboard Cite

Metabolomics experiments can employ non-targeted tandem mass spectrometry to detect hundreds to thousands of molecules in a biological sample. Structural annotation of molecules is typically carried out by searching their fragmentation spectra in spectral libraries or, recently, in structure databases. Annotations are limited to structures present in the library or database employed, prohibiting a thorough utilization of the experimental data. We present a computational tool for systematic compound class annotation: CANOPUS uses a deep neural network to predict 1,270 compound classes from fragmentation spectra, and explicitly targets compounds where neither spectral nor structural reference data are available. CANOPUS even predicts classes for which no MS/MS training data are available. We demonstrate the broad utility of CANOPUS by investigating the effect of the microbial colonization in the digestive system in mice, and through analysis of the chemodiversity of different Euphorbia plants; both uniquely revealing biological insights at the compound class level.DNN is trained on 1.11 million compound structures and does not require any MS/MS data. To train the DNN, we have to simulate a "realistic" probabilistic fingerprint for any given molecular structure, although no MS/MS data for this structure is available. This integration of two machine learning techniques allows CANOPUS to reach high-quality predictions for 1,270 compound classes. Because the predictions are now independent from the availability of MS/MS reference data, CANOPUS can predict compound classes even when there are no MS/MS spectra for training the method. Equally important, it can predict classes for which MS/MS training data is missing for a complete subclass.Uniquely, CANOPUS permits a global over view of the compound classes measured in a biological sample, but also the differences between cohorts at the compound class level

show abstract

Section: Introductionmentioning

confidence: 99%

Classes for the masses: Systematic classification of unknowns using fragmentation spectra

Dührkop

Nothias

Fleischauer

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…The “plan” step involved narrowing the chemical search space by extracting structural information directly from the target mass spectrum. A range of machine-learning methods were proposed to address this step, most of which were aimed at identifying likely substructures of the target molecule. − This is the approach that is routinely applied as part of the NIST 2014/EPA/NIH MS Search . The “generate” step generates candidate chemical structures from within that refined search space.…”

mentioning

confidence: 99%

Computational Prediction of Electron Ionization Mass Spectra to Assist in GC/MS Compound Identification

et al. 2016

View full text Add to dashboard Cite

We describe a tool, competitive fragmentation modeling for electron ionization (CFM-EI) that, given a chemical structure (e.g., in SMILES or InChI format), computationally predicts an electron ionization mass spectrum (EI-MS) (i.e., the type of mass spectrum commonly generated by gas chromatography mass spectrometry). The predicted spectra produced by this tool can be used for putative compound identification, complementing measured spectra in reference databases by expanding the range of compounds able to be considered when availability of measured spectra is limited. The tool extends CFM-ESI, a recently developed method for computational prediction of electrospray tandem mass spectra (ESI-MS/MS), but unlike CFM-ESI, CFM-EI can handle odd-electron ions and isotopes and incorporates an artificial neural network. Tests on EI-MS data from the NIST database demonstrate that CFM-EI is able to model fragmentation likelihoods in low-resolution EI-MS data, producing predicted spectra whose dot product scores are significantly better than full enumeration "bar-code" spectra. CFM-EI also outperformed previously reported results for MetFrag, MOLGEN-MS, and Mass Frontier on one compound identification task. It also outperformed MetFrag in a range of other compound identification tasks involving a much larger data set, containing both derivatized and nonderivatized compounds. While replicate EI-MS measurements of chemical standards are still a more accurate point of comparison, CFM-EI's predictions provide a much-needed alternative when no reference standard is available for measurement. CFM-EI is available at https://sourceforge.net/projects/cfm-id/ for download and http://cfmid.wishartlab.com as a web service.

show abstract

“…Many kinds of structural analysis methods of mass spectra have been represented and developed, such as the pattern recognition technique and structure generation. [1][2][3][4][5][6] In this paper, a chemical structure is regarded as a graph and is described hierarchically by means of not only the structural unit, "atom", but also the intermediate concept, "block". This hierarchical representation simplifies the analyzing procedure and saves processing time.…”

Section: Introductionmentioning

confidence: 99%

Learning system for automatic structural analysis of mass spectra

Nakayama¹,

Fujiwara²

1981

J. Chem. Inf. Comput. Sci.

View full text Add to dashboard Cite

Comparison of various K-nearest neighbor voting schemes with the self-training interpretive and retrieval system for identifying molecular substructures from mass spectral data

Cited by 24 publications

References 15 publications

Classes for the masses: Systematic classification of unknowns using fragmentation spectra

Classes for the masses: Systematic classification of unknowns using fragmentation spectra

Computational Prediction of Electron Ionization Mass Spectra to Assist in GC/MS Compound Identification

Learning system for automatic structural analysis of mass spectra

Contact Info

Product

Resources

About