Better consideration of chemical factors in the exposome requires high-quality chemical information in research articles. Many exposomics resources are based (mostly) on literature mining using name and synonym matching, which can be notoriously prone to errors. In this section, we provide some guidance on what information authors should consider providing, as well as the pros and cons of various choices. Since this Chemical Structure Data template was presented recently to the Journal of Cheminformatics,¹² some of the material in this section overlaps with the previous article.

Authors should consider submitting their chemical structure information with their manuscript as Supplementary Material using the suggested template as comma separated value (CSV; *.csv); or, alternatively, as tab-separated value (TSV; *.tsv) or structure data file (SDF; *.sdf) formats. These formats ensure maximum interoperability between resources and operating systems. The popular XLS(X) format is not truly interoperable (options to save as CSV or TSV are offered), while the extraction of information from PDF format is difficult without introducing errors. The content below describes the CSV/TSV formats, SDF instructions are available elsewhere¹⁷ (however, the SD fields should match the CSV/TSV headers). In our experience, so far CSV often proves most interoperable for the widest audience, although the other formats also have certain advantages.

For CSV/TSV files, the header (first row) indicates the data content of each column; each subsequent row corresponds to a complete chemical record description: chemical structure, chemical names, identifiers, comments, and any other data the authors wish to provide (as additional columns). The interoperable case-insensitive template CSV/TSV column headers (or SDF SD fields) are: SMILES, InChI, and InChIKey for chemical structure; Name and Synonym for chemical names; and Comment for textual comments. Any additional columns headers (e.g., for data, additional identifiers, or desired metadata) are up to the author (e.g., the PubChem_CID identifier header in Figure 2). Note that there may be many Synonym and Comment columns in the file to provide space for more chemical names and metadata, respectively.

The author-submitted template file¹⁸ should contain at least one of the following columns: SMILES, InChI, Name, or InChIKey. The Name column corresponds to a single primary name for the chemical structure. Each Synonym column corresponds to an additional chemical name (one name entry per column). Each Comment column can be added to provide additional text that may be important to the downstream user. Authors can also provide additional CSV/TSV columns (or SDF SD fields) containing information about their chemical substances (with unique, descriptive headers) for additional context. Chemical database identifiers or registry numbers could be included in this manner (as additional columns or fields), or as a Synonym. Note that chemical records indicating chemical structure with only InChIKey or Name will not contain sufficient information to describe a chemical structure; and can only be mapped to existing entries in destination resources. Batch services are available (e.g., from PubChem^7,20 or CompTox^9,21) for authors to add, e.g., SMILES and/or InChI to their records, based upon the Name or other identifiers.

Figure 1 in Schymanski and Bolton¹¹ shows the template file, which is available for download¹⁸ and as Supplementary Material with this article. Figure 2 shows an example submission according to the proposed template, created by sub-setting the “HSDBTPS” dataset of literature-mined and curated transformation products from the Hazardous Substance Data Bank (HSDB) in PubChem.^19,22 This example provides the Name, SMILES, and InChIKey fields as suggested, and an identifier (the PubChem Compound Identifier, CID) as an additional (optional) column (PubChem_CID) with a unique and easily recognizable header that can be processed by other resources as they choose, helping with interoperability.

The advancement of modern science is data driven.^23,24 Providing key data in a ready to use format helps to assist in its reuse in research articles, regulatory reports, or machine-learning data models. Exposomics especially needs access to ready-to-use, high-quality chemical information from individual research articles (e.g., such as the connection of detected chemicals with the disease endpoint investigated or the aggregation of known metabolites of thousands of common chemicals). For instance, HSDB contains metabolites and metabolism information for 3220 chemicals gathered over 40 years, but these are only available as text snippets that need to be matched to chemical structures by synonyms followed by manual curation (initial efforts have covered only 1/100th of this dataset²²). However, as mentioned above, a key challenge in exposomics is to connect chemicals (e.g., of anthropogenic origin, but also endogenous or exogenous chemicals) that are associated with exposures with their biological response. Since metabolism is the most dynamic of the biological responses, and metabolites per definition fall into the same molecular mass category as many anthropogenic chemicals of concern, a key gap in exposomics knowledge is the connection between chemicals and their metabolites. The efforts of many will be needed to help fill this knowledge gap and the timing could not be better for exposomics with several recent studies emerging using in vitro enzymes to investigate parent–metabolite relationships of drugs and other relevant chemicals.^25,26

The Transformations template provided here has been designed on the basis of recent efforts to fill the gaps of transformation products in PubChem using literature data,²⁷ in collaboration with the NORMAN Suspect List Exchange (NORMAN-SLE).^28-30 Several datasets from a variety of sources have now been processed. Transformations from the NORMAN-SLE, where S## refers to the list number, followed by the list code, include: S60 SWISSPEST19,^31,32 S66 EAWAGTPS,^33,34 S68 HSDBTPS,^21,22 S73 METXBIODB,^35,36 S74 REFTPS,³⁷ S78 SLUPESTTPS,^38,39 S79 UACCSCEC,^40,41 and S81 THSTPS⁴² (list available from https://git-r3lab.uni.lu/eci/pubchem/-/raw/master/annotations/tps/Transformation_Datasets.txt). Of these, MetXBioDB also contains enzyme information, while the rest are primarily environmental data. Figure 3 shows an example “environmental” dataset compiled from several of these lists, using the proposed template. In addition to the NORMAN-SLE datasets, a dataset of more than 1200 transformations from ChEMBL⁴³ has also been added, including enzyme, gene, and protein information (where available). An example of Transformations with more biological information available is given in Figure 4.

Information about both the predecessor (parent/precursor) and successor (transformation product/metabolite) must be given for a valid transformation. The template can accept at least one of Name, SMILES, or PubChem CID for each, where SMILES or CID is preferred, and SMILES will be the most interoperable. Note that these need not be consistent—for instance, it is possible to provide SMILES of the successor and a CID of the predecessor if a Name or CID is not available for the successor. It is preferable to give two fields, Figure 3 shows the example of Name and CID, while Figure 4 an example of SMILES and Name (top panel on each figure).

If available, a brief description of the transformation is useful and can be provided in the “Transformation” field (top panel, Figures 3 and 4). Short, informative descriptions are preferred; the current entries have been either extracted automatically from existing datasets or entered manually. In the future, it may be possible to provide some guidance via an ontology as the public dataset grows to improve the machine readability. Similarly, if information on the biosystem is available (i.e., where the transformation takes place), this can be included in the Biosystem column (see Figures 3 and 4 for examples).

For datasets with biological information, this can be provided (optionally) in the Enzyme, Gene_ID, and Protein_ID columns. At this stage, the template allows flexible input (see Figure 4 for examples) but recommend Enzyme are provided as either: Enzyme Commission (EC) number,^45–47 such as “EC 2.3.2.23”; gene symbol, such as “CYP1A1”; or as enzyme names, such as “Aryl hydrocarbon hydroxylase.” The Gene_ID is expected to be an NCBI Gene⁴⁸ ID, such as “1543.” The Protein_ID is expected to be either an NCBI Protein⁴⁹ accession, such as “NP_059488.2” or an UniProt identifier,⁵⁰ such as “P08684.” If multiple entries for Enzyme, Gene_ID, and Protein_ID are provided, they should be separated by a “pipe” symbol (“|”) or provided as new rows.

Finally, the Reference_ID and Reference_Description columns provide the opportunity to credit the original sources of the information. Reference_ID entries should be either PubMed identifiers⁵¹ (PMIDs) or Digital Object Identifiers⁵² (DOIs), preceded with “PMID:” or “DOI:”, respectively, for easy recognition, and separated by a “pipe” (“|”) if multiple IDs exist (they can be mixed—for example, “PMID:33929905|DOI:10.1186/s13321-018-0324-5”). The Reference_Description can be used to provide a free text form of the reference, to describe the data source (if no PMID/DOI available) or to describe evidence of the transformation. Only Reference_ID can be processed automatically. Again, see Figures 3 and 4 and the Transformations template⁴⁴ for examples.

So far, about 6000 Transformations have been processed using these templates, from nine different sources (many of these being composite data from several sources themselves, including ChEMBL,⁴³ MetXBioDB,³⁵ and REFTPS³⁷). The Transformations are being integrated into current computational mass spectrometry workflows (such as patRoon⁵³ and as documented in Krier et al.²²) and are openly available for all. The summarized files are likewise available for comprehensive efforts such as BioTransformer³⁶ to add this new data to their training set (MetXBioDB³⁵ is the library behind BioTransformer) and likewise improve predictions. Overall, FAIR transformations data will greatly support exposomics, and discussions to extend these templates into fields with formal ontologies and/or other formats such as mzTab^54,55 in the future are welcomed. As demonstrated in Figures 5 and 6, one can see the benefits of arranging data in FAIR templates. Figure 5 is an example of a resulting Transformation entry in PubChem, while Figure 6 can be created automatically in CDK Depict using simple code in R to create annotated reaction SMILES from the fields shown in Figure 3 only.

Exposomics is a data-driven science and vast quantities of information will be needed for it to be successful. By making the output of exposomics research available in a more machine-readable way, we can accelerate our progress and rise to the challenge. The templates provided here are a means to make primary outputs FAIR (Findable, Accessible, Interoperable, and Reusable). When authors provide this content as Supplementary Material, it can be readily accessed and utilized, ideally without human intervention. When the journal interlinks these Supplementary Material files with the article DOI and associated metadata, other resources can rapidly find and integrate this content and provide enhanced services for the entire community. Improving the FAIRness of Supplementary Material greatly decreases the effort to combine and aggregate information between papers and improves the correctness of the information over text-mining-based approaches. It also greatly enhances the visibility of the individual works and research outputs. As a young scientific discipline, the exposome should learn from its closely related “elder” disciplines. Genomic approaches gained incredible traction due to the widely encouraged and eventually mandated sharing of information. Let us take these lessons to heart and advance together as a field. We need to share information—and lots of it—to help make sense of the exposome. The use of these facile, ready-to-use templates will help advance exposomics by contributing vital information to complete the exposomics “puzzle.”