Search

The extraction of chemical information from documents is a demanding task in cheminformatics due to the variety of text and image-based representations of chemistry. The present work describes the extraction of chemical compounds with unique chemical structures from the open access CORE (COnnecting REpositories) and Google Patents full text document repositories. The importance of structure normalization is demonstrated using three open access cheminformatics toolkits: CDK, RDKit and OpenChemLib. Each toolkit was used for structure parsing, normalization and subsequent substructure searching, using SMILES as structure representations of chemical molecules. Per- and polyfluoroalkyl substances (PFAS) were chosen as a case study to perform the substructure search, due to their high environmental relevance, their presence in both literature and patent corpuses, and the current lack of community consensus on their definition. Three different structural definitions of PFAS were chosen to highlight the implications of various definitions from a cheminformatics perspective. Since CDK, RDKit and OpenChemLib implement different criteria and methods for SMILES parsing and normalization, different numbers of parsed compounds were extracted, which were then evaluated using the three PFAS definitions. A comparison of these toolkits and definitions is provided, along with a discussion of the implications for PFAS screening and text mining efforts in cheminformatics. Finally, the extracted PFAS (~1.7 M PFAS from patents and ~27K from CORE) were compared against various existing PFAS lists and are provided in various formats for further community research efforts.

Diese Arbeit umfasst Methoden für die computergestützte Auswertung von Massenspektrometriedaten mithilfe derer die Identifikation von niedermolekularen Substanzen verbessert wird. Hierfür wird die kombinatorische Fragmentierung durch die Kombination von unterschiedlichen Strategien und Informationsquellen erweitert. Die entwickelten Ansätze ermöglichen die Einbindung von Daten verschiedener experimenteller Methoden, wie zum Beispiel dem Wasserstoff-Deuterium-Austausch. Diese zusätzlichen Informationen helfen, die Anzahl der korrekt identifizierten Moleküle zu erhöhen. Zudem werden statistische Methoden verwendet, um Substrukturannotationen zu lernen und die Identifikationleistung bestehender computergestützter Methoden zu übertreffen. Daher präsentieren wir eine automatisierte Pipeline, die bereits in verschiedenen Softwareanwendungen, sowie zahlreichen Metabolomik- und Umweltstudien zum Einsatz kam, um die Identifikation von niedermolekularen Substanzen zu unterstützen., In this thesis, we develop methods for computational mass spectrometry to improve the identification of small molecules based on tandem mass spectra. Therefore, we combine different strategies and information sources to extend combinatorial fragmentation for the evaluation of molecular candidates. The developed algorithmic methods enable the usage of data from additional experiments, such as hydrogen deuterium exchange mass spectrometry. Thus, the included information show an increase in the number of correctly identified molecules. Furthermore, statistical methods used to learn substructure annotations from reference spectra help to outperform existing computational methods. We present a new and refactored automated pipeline, initially known as MetFrag, that have been successfully applied in various software solutions and studies to support the identification of small molecules in metabolomics and environmental research.

Liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) is increasingly popular for the non-targeted exploration of complex samples, where tandem mass spectrometry (MS/MS) is used to characterize the structure of unknown compounds. However, mass spectra do not always contain sufficient information to unequivocally identify the correct structure. This study investigated how much additional information can be gained using hydrogen deuterium exchange (HDX) experiments. The exchange of “easily exchangeable” hydrogen atoms (connected to heteroatoms), with predominantly [M+D]+ ions in positive mode and [M-D]− in negative mode was observed. To enable high-throughput processing, new scoring terms were incorporated into the in silico fragmenter MetFrag. These were initially developed on small datasets and then tested on 762 compounds of environmental interest. Pairs of spectra (normal and deuterated) were found for 593 of these substances (506 positive mode, 155 negative mode spectra). The new scoring terms resulted in 29 additional correct identifications (78 vs 49) for positive mode and an increase in top 10 rankings from 80 to 106 in negative mode. Compounds with dual functionality (polar head group, long apolar tail) exhibited dramatic retention time (RT) shifts of up to several minutes, compared with an average 0.04 min RT shift. For a smaller dataset of 80 metabolites, top 10 rankings improved from 13 to 24 (positive mode, 57 spectra) and from 14 to 31 (negative mode, 63 spectra) when including HDX information. The results of standard measurements were confirmed using targets and tentatively identified surfactant species in an environmental sample collected from the river Danube near Novi Sad (Serbia). The changes to MetFrag have been integrated into the command line version available at http://c-ruttkies.github.io/MetFrag and all resulting spectra and compounds are available in online resources and in the Electronic Supplementary Material (ESM). ISSN:1618-2650 ISSN:1618-2642

Figure S1 - Weight Parameter Scan for the test dataset. (PDF 767 kb)

Figure S2 - Maximum spectral similarities. (PDF 196 kb)

Per- and polyfluoroalkyl substances (PFASs) are a large and diverse class of chemicals of great interest due to their wide commercial applicability, as well as increasing public concern regarding their adverse impacts. A common terminology for PFASs was recommended in 2011, including broad categorization and detailed naming for many PFASs with rather simple molecular structures. Recent advancements in chemical analysis have enabled identification of a wide variety of PFASs that are not covered by this common terminology. The resulting inconsistency in categorizing and naming of PFASs is preventing efficient assimilation of reported information. This article explores how a combination of expert knowledge and cheminformatics approaches could help address this challenge in a systematic manner. First, the “splitPFAS” approach was developed to systematically subdivide PFASs (for eventual categorization) following a CnF2n+1–X–R pattern into their various parts, with a particular focus on 4 PFAS categories where X is CO, SO2, CH2 and CH2CH2. Then, the open, ontology-based “ClassyFire” approach was tested for potential applicability to categorizing and naming PFASs using five scenarios of original and simplified structures based on the “splitPFAS” output. This workflow was applied to a set of 770 PFASs from the latest OECD PFAS list. While splitPFAS categorized PFASs as intended, the ClassyFire results were mixed. These results reveal that open cheminformatics approaches have the potential to assist in categorizing PFASs in a consistent manner, while much development is needed for future systematic naming of PFASs. The “splitPFAS” tool and related code are publicly available, and include options to extend this proof-of-concept to encompass further PFASs in the future.

Presentation at EPA's Non-Targeted Analysis Collaborative Trial (ENTACT) Workshop Aug 2018

Table S2 - Notation summary (Scores). (PDF 70.4 kb)

Figure S3 - Rankings of the correct candidates (test) vs. max. spectral similarity. (PDF 204 kb)

Background Molecule identification is a crucial step in metabolomics and environmental sciences. Besides in silico fragmentation, as performed by MetFrag, also machine learning and statistical methods evolved, showing an improvement in molecule annotation based on MS/MS data. In this work we present a new statistical scoring method where annotations of m/z fragment peaks to fragment-structures are learned in a training step. Based on a Bayesian model, two additional scoring terms are integrated into the new MetFrag2.4.5 and evaluated on the test data set of the CASMI 2016 contest. Results The results on the 87 MS/MS spectra from positive and negative mode show a substantial improvement of the results compared to submissions made by the former MetFrag approach. Top1 rankings increased from 5 to 21 and Top10 rankings from 39 to 55 both showing higher values than for CSI:IOKR, the winner of the CASMI 2016 contest. For the negative mode spectra, MetFrag’s statistical scoring outperforms all other participants which submitted results for this type of spectra. Conclusions This study shows how statistical learning can improve molecular structure identification based on MS/MS data compared on the same method using combinatorial in silico fragmentation only. MetFrag2.4.5 shows especially in negative mode a better performance compared to the other participating approaches. Electronic supplementary material The online version of this article (10.1186/s12859-019-2954-7) contains supplementary material, which is available to authorized users.

International audience

Background: The fourth round of the Critical Assessment of Small Molecule Identification (CASMI) Contest ( www.casmi-contest.org ) was held in 2016, with two new categories for automated methods. This article covers the 208 challenges in Categories 2 and 3, without and with metadata, from organization, participation, results and post-contest evaluation of CASMI 2016 through to perspectives for future contests and small molecule annotation/identification. Results: The Input Output Kernel Regression (CSI:IOKR) machine learning approach performed best in "Category 2: Best Automatic Structural Identification - In Silico Fragmentation Only", won by Team Brouard with 41% challenge wins. The winner of "Category 3: Best Automatic Structural Identification - Full Information" was Team Kind (MS-FINDER), with 76% challenge wins. The best methods were able to achieve over 30% Top 1 ranks in Category 2, with all methods ranking the correct candidate in the Top 10 in around 50% of challenges. This success rate rose to 70% Top 1 ranks in Category 3, with candidates in the Top 10 in over 80% of the challenges. The machine learning and chemistry-based approaches are shown to perform in complementary ways. Conclusions: The improvement in (semi-)automated fragmentation methods for small molecule identification has been substantial. The achieved high rates of correct candidates in the Top 1 and Top 10, despite large candidate numbers, open up great possibilities for high-throughput annotation of untargeted analysis for "known unknowns". As more high quality training data becomes available, the improvements in machine learning methods will likely continue, but the alternative approaches still provide valuable complementary information. Improved integration of experimental context will also improve identification success further for "real life" annotations. The true "unknown unknowns" remain to be evaluated in future CASMI contests. Graphical abstract .

Additional file 1. This file contains additional content (methods, results and selected spectra) to complement the manuscript. See PDF for details.

Additional file 9. Top 10 ranks with ChemSpider on the Orbitrap XL Dataset The results were obtained with MetFrag2.2 formula query and the inclusion of references and retention time. Each small dot shows the number of first ranks with a given combination of weights. The larger dots show the best result (471 in the top 10), 90 th percentile (460), median (450), 10 th percentile (404) and worst result (223).

Liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) is increasingly popular for the non-targeted exploration of complex samples, where tandem mass spectrometry (MS/MS) is used to characterize the structure of unknown compounds. However, mass spectra do not always contain sufficient information to unequivocally identify the correct structure. This study investigated how much additional information can be gained using hydrogen deuterium exchange (HDX) experiments. The exchange of “easily exchangeable” hydrogen atoms (connected to heteroatoms), with predominantly [M+D]+ ions in positive mode and [M-D]− in negative mode was observed. To enable high-throughput processing, new scoring terms were incorporated into the in silico fragmenter MetFrag. These were initially developed on small datasets and then tested on 762 compounds of environmental interest. Pairs of spectra (normal and deuterated) were found for 593 of these substances (506 positive mode, 155 negative mode spectra). The new scoring terms resulted in 29 additional correct identifications (78 vs 49) for positive mode and an increase in top 10 rankings from 80 to 106 in negative mode. Compounds with dual functionality (polar head group, long apolar tail) exhibited dramatic retention time (RT) shifts of up to several minutes, compared with an average 0.04 min RT shift. For a smaller dataset of 80 metabolites, top 10 rankings improved from 13 to 24 (positive mode, 57 spectra) and from 14 to 31 (negative mode, 63 spectra) when including HDX information. The results of standard measurements were confirmed using targets and tentatively identified surfactant species in an environmental sample collected from the river Danube near Novi Sad (Serbia). The changes to MetFrag have been integrated into the command line version available at http://c-ruttkies.github.io/MetFrag and all resulting spectra and compounds are available in online resources and in the Electronic Supplementary Material (ESM)., Analytical and Bioanalytical Chemistry, 411 (19), ISSN:1618-2650, ISSN:1618-2642

Additional file 5. Additional MetFrag results and data (Figures S5â S8, Table S4).

Additional file 3. CompTox Chemistry Dashboard search interfaces (Figures S1â S4).

Additional file 3. CompTox Chemistry Dashboard search interfaces (Figures S1â S4).

Additional file 5. Additional MetFrag results and data (Figures S5â S8, Table S4).