Quantitative Structure Use Relationships: Highlights from a technical summit meeting

Abstract

The Quantitative Structure Use Relationship (QSUR) Summit, held on November 2–4, 2022, focused on advancing the development, refinement, and use of QSURs to support chemical substance prioritization and risk assessment and mitigation. QSURs utilize chemical structures to predict the function of a chemical within a formulated product or an industrial process. This presumed function can then be used to develop chemical use categories or other information necessary to refine exposure assessments. The invited expert meeting was attended by 38 scientists from Canada, Finland, France, the UK, and the USA, representing government, business, and academia, with expertise in exposure science, chemical engineering, risk assessment, formulation chemistry, and machine learning. Workshop discussions emphasized the importance of collection and sharing of data and quantification of relative chemical quantities to progress QSUR development. Participants proposed collaborative approaches to address key challenges, including mechanisms for aggregating information while still protecting proprietary product composition and other confidential business information. Discussions also led to proposals for applications beyond exposure and risk modeling, including sustainable formulation discovery. In addition, discussions continue to construct, conduct, and circulate case studies tied to various specific problem formulations in which QSURs supply or derive information on chemical functions, concentrations, and exposures.

1. Introduction

Reliable exposure quantification is required for greater confidence in chemical safety and risk assessment and chemicals management. Understanding where, when and how chemicals are used in commerce is critical for estimating exposures but is currently a key source of uncertainty for many chemicals (Comiskey et al., 2015). The United States Environmental Protection Agency (US EPA) developed the concept of Quantitative Structure Use Relationships (QSURs), wherein the chemical structure is used to predict a chemical’s function within a formulated product or an industrial process (Phillips et al., 2017; Isaacs et al., 2022). This function can then be used to estimate other needed information to replace defaults (e.g., to refine weight fractions within a given product type or category (Isaacs et al., 2016) or releases into the environment from product manufacturing and use (Li et al., 2021)). A QSUR Summit (held November 2–4, 2022) focused on advancing development, refinement, and use of QSURs to support chemical substance prioritization, risk assessment, and mitigation. Scientists representing government, business, and academia from Canada, Finland, France, the UK, and the USA participated in the workshop and provided expertise in exposure science, chemical engineering, risk assessment, formulation chemistry, and machine learning. The workshop focused on facilitating interdisciplinary connections and knowledge sharing to advance exposure science, enhance QSUR development and application, and identify future research areas.

2. Current status of QSURs

QSUR predictions aim to improve the accuracy of exposure assessment by bridging gaps in the public availability of chemical use information. The information gaps arise from challenges in accessing relevant chemical reporting, use, or release data or making these data public in a way that protects commercial interests; contributing factors are detailed in Supplementary Information (SI).

US EPA’s Office of Research and Development (ORD) has led QSUR development. Existing QSURs for chemical function have focused on consumer products, with current activities expanding into industrial sector uses. Results of QSUR predictions can be found in US EPA’s ExpoCast information on the CompTox dashboard (https://comptox.epa.gov/dashboard/); QSUR algorithms are also available on an R platform (https://github.com/HumanExposure/qsur).

EPA has demonstrated the utility of using QSURs for estimating the likelihood of presence within a consumer product (Isaacs et al., 2016), determining weight fraction within products (Isaacs et al., 2016), interpreting Non-Targeted Analysis (NTA) data (Phillips et al., 2018; Lowe et al., 2021), and identifying potential functional replacements in alternatives assessment (Phillips et al., 2017). QSURs have also been used to estimate chemical emissions across the lifecycle. Table 1 contains a list of models, databases, and generic formulation documents (i.e., frame formulations¹) relevant to QSUR application and/or development. Analyses have shown that current QSURs generally work well for estimating the weight fraction of single-function chemicals in a product, but performance is more variable for substances with multiple functional uses in products (Gouin, 2022). Addressing multi-function chemicals is important, as chemicals often find new applications once placed on the market. Developing QSURs for multi-function substances needs to consider a diversity of functions for one substance depending on the product context, and may require co-consideration of aspects such as a combination of product category and function. Obtaining representative datasets for model development is therefore key.

Table 1.

Models and data resources (summary).

Part 1: Exposure Models
Model Name and Description	Relevance to QSURs
CiP-CAFÉ  Provided with user-supplied chemical tonnages (annual production or consumption volume in metric tons per year), CiP-CAFE (version 2.0) calculates chemical flows between main life cycle stages and waste disposal practices, as well as rates of emissions therefrom. Availability: Free; Reference(s) and/or Weblink: www.eas-e-suite.com	Uses QSURs to determine chemical use (function) in high-throughput applications.
PROTEX-HT Consolidates various mass-balance models to simulate chemical emissions throughout their life-cycle, indoor and outdoor fate and transport, food web bioaccumulation, and aggregate exposures and potential risks to representative humans and ecological receptors. PROTEX-HT integrates the CiP-CAFE Ver.2.0 model for steady-state (time invariant) simulations. Availability: Free; Reference(s) and/or Weblink: www.eas-e-suite.com	Uses QSURs to determine chemical use (function) in high-throughput applications.
SEEM3  Systematic Empirical Evaluation of Models (SEEM) framework. Available predictors are combined into a consensus Bayesian regression meta-model for median population exposure via comparison with human biomonitoring data. SEEM is a “meet-in-the-middle” approach in which predictions of chemical intake rates from “forward” models are compared with rates inferred from the NHANES biomarker data using “reverse” models that attempt to reconstruct exposures. These could be further informed by QSUR predictions. Availability: Free; Reference(s) and/or Weblink: https://comptox.epa.gov/dashboard/	The SEEM3 framework includes simple QSURs to determine chemical use and exposure pathways. These QSURS could be refined in the future.
SHEDS-HT  Aggregate population exposure for chemicals, based on consumer product use and/or food intakes. Uses ~300 hierarchical consumer product categories. Availability: Free; Reference(s) and/or Weblink: https://github.com/HumanExposure/SHEDSHTRPackage	QSURs could be updated to predict function and/or weight fractions for specific consumer product categories.
Part 2: Databases and Resources	Relevance to QSURs
Cosmetics Ingredients Review  Parameters relevant to QSURs: chemicals evaluated typically include weight fraction for some (or a number of) product types. Availability: Free; Reference(s) and/or Weblink: https://www.cir-safety.org/ingredients	Includes information on function and the types of products in which a substance may typically be used.
CPDat (Chemicals and Products Database)  Parameters relevant to QSURs: chemical name and function(s). Current QSUR predictions for thousands of chemicals. Availability: Free; Reference(s) and/or Weblink: https://www.epa.gov/chemical-research/chemical-and-products-database-cpdat) CPDat can be interactively explored via EPA’s ChemExpo Knowledgebase (ChemExpo, https://comptox.epa.gov/chemexpo/	Provides training data for QSURs - information on chemical function in products and processes curated from public sources.
Euromonitor  Commercial database including information on chemical presence in product types and sales volume. Database includes US information as well as for EU. Availability: Fee for use; Reference(s) and/or Weblink: https://www.euromonitor.com/	Information on substance presence and volume can be used to benchmark QSURs- see Gouin et al., 2022.
Environment and Climate Change Canada (ECCC) ERC2  Very detailed exposure and hazard information for 12200 organic chemicals by CAS including QSUR and known functional uses. Availability: Free; Reference(s) and/or Weblink: https://www.canada.ca/en/environment-climate-change/services/evaluating-existing-substances/science-approach-document-ecological-risk-classification-organic-substances-erc2.html	Uses QSURs from US EPA models.
Household and Commercial Products Ingredient Database  Chemical name, function, product category for definitions of the chemicals used specifically in household and automotive care consumer products. Availability: Limited free access; full access available through subscription; Reference(s) and/or Weblink: https://www.thehcpa.org/resources/ingredient-dictionary/	Can link function to product category, chemical name.
REACH Registration Database on Substances  Each of the 23,000 substances (identified by name, EC number, CAS number, trade name, representative structures, composition by constituents) is associated with a high level use profile in structured format, consisting of: life-cycle stages and main user groups (consumer and/or professional and/or industrial), technical function(s), chemical product [mixture] categories, article categories (if relevant), environmental release categories, process categories, use name (freetext) and use description (freetext). The use profile does not include data on substance amounts by area of use or on concentration in formulated products. For each substance also a data-set on basic chemical-physical properties is available. Availability: substance function and substance composition by constituents is not publicly disseminated; for single substances public access to some other parameters may also be restricted. Reference(s) and/or Weblink: https://echa.europa.eu/information-on-chemicals/registered-substances/	Can link substance to representative structure and physical-chemical parameters, to intended substance function and expected occurrence in products (however usually no weight fractions available); function and product occurrence are based on knowledge of substance manufacturers and importers.
Substances in Preparations in Nordic Countries (SPIN)  Searchable by substance. Provides use volumes by use categories and includes function categories. Information is by country and year also. Availability: Free; Reference(s) and/or Weblink: http://spin2000.net/	Can link substance use and function, and also general product categories. Can also look at trends over time.
OECD Emission Scenario Documents (ESDs)  ESDs provide information on the sources, use patterns, and potential release pathways of chemicals in a wide spectrum of products. ESDs present standard approaches for estimating the environmental releases of and exposures to chemicals in products. There are currently 40 published ESDs. Availability: Free; Reference(s) and/or Weblink: https://www.oecd-ilibrary.org/environment/series-on-emission-scenario-documents_23114606	Weight fraction included in some cases. Potential use in developing or benchmarking weight fraction estimates based upon QSURs.
REACH Use Map Library  Industry sector information on the uses and the conditions of use to be expected for certain types of formulated products (mixtures) or functional substances (e.g. solvents), providing also input parameters for Tier 1 exposure estimates. Availability: Free; Reference(s) and/or Weblink: https://www.echa.europa.eu/csr-es-roadmap/use-maps/use-maps-library	May include upper bound for weight fraction by formulated product categories. Potential use in developing or benchmarking weight fraction estimates based upon QSURs.
Part 3: Frame Formulations: Generic formula by product type, typically indicating functional ingredients and weight fraction ranges or maximum values	Relevance to QSURs
Cleaning Products: RIVM Cleaning Products Fact Sheet  Generic formula by product type, typically indicating functional ingredients and weight fraction ranges or maximum values. Product categories included: Abrasives, fabric conditioners, liquid hand dishwashing detergent, all-purpose cleaners, floor cleaning protecting products, machine dishwashing detergent, bathroom cleaners, glass cleaners, metal cleaners, carpet cleaners, laundry detergent powders, oven cleaners, dishwashing machine rinse aids, laundry liquid products, toilet cleaners. Availability: Free; Reference (s) and/or Weblink: https://www.rivm.nl/bibliotheek/rapporten/2016-0179.pdf. Full list of RIVM Fact Sheets can be found at: https://www.rivm.nl/en/consexpo/fact-sheets	Assist in QSUR development and benchmarking, based upon substance function can use to predict likely presence in a formulated product and potentially weight fraction.
Cleaning Products: Schill + Seilacher Household and I & I Formulation Guide  Product categories included: car wash concentrate, handwashing dish detergent, oven cleaner, car wash shampoo, hard surface cleaners, shoe refresher spray, dishwasher rinse aid, high-pressure car cleaner, sponge impregnation, fabric refresher, laundry detergent, tile and bath cleaners, floor cleaner, laundry handwash detergent, toilet cleaners, glass cleaner, laundry softener. Availability: Free; Reference(s) and/or Weblink: https://www.schillseilacher.de/fileadmin/user_upload/Produkte/Kosmetik/Formulation_GuideHousehold_2020.pdf	Assist in QSUR development and benchmarking, based upon substance function can use to predict likely presence in a formulated product and potentially weight fraction.
Cosmetic: Cosmetics Products Notification Panel  Product categories included: deodorant/antiperspirant, make-up, shaving, hair & scalp, mouthwash, skin care, hair colouring, nail care, skin cleansers, hair removal, nail varnish, sun products & self-tanning, hair styling, perfumes, tooth care. Availability: Free, user must create account; Reference(s) and/or Weblink: https://webgate.ec.europa.eu/cpnp/resources/ff/FF-2013-EN-TRA-00.pdf	Assist in QSUR development and benchmarking, based upon substance function can use to predict likely presence in a formulated product and potentially weight fraction.
Paints: RIVM Paints Fact Sheet  Product categories include: 2-component, spray, waterborne emulsion, high solid, waterborne dispersion, waterborne wall, solvent rich. Availability: Free; Reference(s) and/or Weblink: https://www.rivm.nl/bibliotheek/rapporten/320104008.pdf. Full list of RIVM Fact Sheets can be found at: https://www.rivm.nl/en/consexpo/fact-sheets	Assist in QSUR development and benchmarking, based upon substance function can use to predict likely presence in a formulated product and potentially weight fraction.
Gosselin et al Clinical Toxicology of Commercial Products, 5th Williams and Wilkins, Baltimore, MD, 1984  General formulations for many categories, including: adhesives, deicers, personal care products, automotive products, deodorizers, pesticides, caulking, lubricants, pet care, cleaners, paints, pharmaceuticals. Availability: out of print, but may be available in scientific libraries or through purchase; Reference(s) and/or Weblink: https://onlinelibrary.wiley.com/doi/epdf/10.1002/jps.2600741037	Note this reference is now approx. 40 years old; information is dated therefore likely of limited utility for predicting likely presence in a formulated product and potentially weight fraction.

3. Perspectives from regulatory science programs

Representatives of regulatory branches of Health Canada (HC), Environment and Climate Change Canada (ECCC), the European Chemicals Agency (ECHA), and US EPA’s Office of Pollution Prevention and Toxics (OPPT) shared their perspectives on QSUR applications. All indicated their interest in possible future QSUR applications, particularly at a screening level. ECCC has used QSURs in their ecological prioritization approaches to hypothesize functional mixtures for future cumulative assessment (ECCC, 2022), and HC has developed QSURs for several industrial product categories, supplementing the work of EPA. ECHA demonstrated several cases where they could envision QSUR use would be helpful to complement the current use-category reporting conducted substance manufacturers in the EU. In addition, ECHA indicated that their data sets might be helpful for QSUR development, a suggestion borne out by the Canadian use of ECHA data to develop preliminary industrial QSURs. USEPA-OPPT indicated that there might be potential for QSURs to be used as a tool in its programs; however, they would need to understand the reliability and representativeness of such information within the context of TSCA (Toxic Substances Control Act).

4. Refinement of QSURs and exposure and risk assessment applications and data sharing

QSUR model evaluation is ongoing and involves comparing QSUR predictions with reported chemical use data. Technologies and expertise outside of exposure science (e.g., machine learning, product formulation engineering) have been applied in QSUR development to some degree, and continued consideration and application will assist in further QSUR development. Data curation remains an integral part of QSUR development. Understanding which product formulations remain stable over time and which are updated more frequently may help prioritize refinement efforts (i.e., rather than revisiting a well-characterized and time-stable product category, resources may be directed to additional or changing product categories and/or new chemical structures).

Next-generation QSURs are being refined to predict functional use within specific commercial sectors and/or product categories and can then be integrated into an overall tiered exposure and risk assessment approach. US EPA-ORD described a three-tiered QSUR framework under development: the first tier predicts chemical function based on structure, the second tier predicts function within the context of use-sector (e.g., consumer, industrial or professional use), and the third tier improves predictions through consideration of specific product or industrial use category (e.g., cosmetics, adhesive manufacturing). The refined QSUR predictions will still be at a generic level, i.e., they are not specific to individual products on the market but have broader application.

A common theme was that QSUR predictions are only as good as the data they are based upon; additional data would help improve existing QSURs and expand the domain of QSUR application. Additional data not yet in the public domain are needed to expand the coverage of product types and to address additional chemistries. Such data may include proprietary or confidential information, hence innovative approaches for access would be needed. Once developed, QSURs can be applied to support evaluations of thousands of chemicals and can be a useful tool in High-Throughput exposure analyses.

Examples of successful approaches for sharing proprietary data were discussed. These included the federated data sharing MELLODDY project (MachinE Learning Ledger Orchestration for Drug DiscoverY, 2023) and using a third-party data aggregator approach (i.e., RIFM-Research Institute for Fragrance Materials, 2023; see SI for more discussion of data sharing). Opportunities must be catalyzed to continue data and information exchange and knowledge sharing across the regulatory science community.

5. QSUR applications beyond exposure and risk modeling

QSURs have been applied in other arenas, such as alternative assessment and NTA applications. In alternative assessments, one of the objectives is to is to adjust compositions to use less toxic alternatives, while maintaining the formulated product’s overall performance. Use of QSURs to support the evaluation and design of safer alternatives could include identifying potential alternatives with similar function and enabling rapid comparison of chemistries with similar functions (see Phillips et al. (2017)). In the NTA area, QSURs have been used to help corroborate tentative chemical identification in products by assessing if the identified structures may have a function consistent with what might be expected for product components (Phillips et al., 2018; Lowe et al., 2021). QSUR predictions are also being evaluated for inclusion as evidence in workflows that attempt to identify specific chemical structures associated with unknown masses in NTA studies of commercial products.

Summit discussions led to the suggestion of several potential focus areas for improvement of QSURs and many ideas for additional QSUR applications (summarized in Table 2); these suggestions support the utility of further QSUR development.

Table 2.

Opportunities for further development of QSURs and suggested additional applications of QSURs.

Opportunities for further development of QSURs
Development Needs and Immediate Applications	• Expansion of domain of applicability via incorporation of industrial chemical use pattern information into models and databases • Increased sensitivity and specificity for multifunction substances • Development of models that identify use within different sectors or products, i.e., industrial solvents and consumer solvents, rather than just solvents • Adaptation of OECD functional use categories • Development of databases or libraries of QSUR results for thousands of known chemical structures • Systematic parameterization of existing high-throughput exposure models for chemical libraries using available QSURs • Incorporation of QSURs into NTA candidate identification workflows
Data and Information Needs	• Development of systematic approaches for mining existing text-based sources to extract relevant data for QSUR development • Identification of sources of training data for poorly characterized industrial or commercial sectors • Improved identification of true negative data for model development • Identification or development of data for improved external validation of QSUR predictions (including reported information from government bodies) • Expansion and refinement of data curation approaches (curation errors can result in poor models) • Further harmonization of use categories (including industry and product categories) across industries, sectors, and geographies • Improved communication and exchange of models and data
Case Studies	• Define case studies that can be used to prioritize efforts to enhance QSURs.  ∘ Case studies should use indicator chemicals that can illustrate applications for related substances or for classes of similar substances  ∘ A problem formulation framework built around QSURs could provide an organizing principle for case studies. • Case study ideas discussed included:  ∘ Use of QSURs to direct data collection for existing chemicals or rule out uses for new chemicals  ∘ Refinement of PROTEX-HT with 2nd generation EPA QSURs under development; compare model predictions to monitoring databases to identify areas of agreement/disagreement, and identify reasons (models and data are not aligned, database curation, QSUR adjustment, etc.) to identify areas for further study
Suggested additional applications of chemical function as an organizing principle and associated QSURs
Product Design and Alternatives Assessment	• Support improved understanding of the role of specific molecules within existing formulations • Support development of improved frame formulations • Identify new chemicals as potential substitutes; identify potential for repurposing existing chemicals • Improve understanding of the link between structure and function via analysis of predictive features • Identify new uses of existing formulation components in a high-throughput manner (rapid formulation discovery) • Organization/expansion of dictionaries for formulators
Regulatory Assessments	• Expand groupings in risk assessment to consider end-use in addition to structure; utilize as a tool to move away from a one chemical at a time approach • Identify analog exposure data for data-poor chemicals • Support prioritization and screening via improved understanding of use and exposure potential • Help characterize articles and complex mixtures in support of product safety assessments • Identify additional uses and exposures for new chemicals (defining scope of needed exposure assessments) • Provide a bridge between chemical groupings for exposure and hazard assessments • Support identification of exposure source associated with new chemicals found in NTA studies of environmental media (inform risk management/mitigation activities) • Inform chemocentric knowledge graphs for organizing exposure pathway information (analogous to methods used for organizing biological knowledge)
Collaboration, Data Sharing and Communication	• Support development of new OECD functional use categories (identify functional spaces that are not currently defined) • Facilitate communication among disparate groups (industry formulators, regulators, risk/exposure scientists) • Support organization and reporting of existing knowledge on use and quantity • Support communication of exposure concepts and research needs, including to non-scientists and students • Organize exposure data for sharing across models and frameworks

6. Path forward

Advancing QSURs requires a sustained effort to enhance collaborations across the regulatory science community to bring exposure scientists together with experts in product development and formulation, materials science, chemical manufacturing, product stewardship, and regulatory frameworks. To continue advancing QSURs, the workshop participants grouped actions, both near-term and longer-term, into the six areas listed below, which are under active discussion:

1. Collaboration – Identify key partners to formulate objectives collectively and to work together (or in complementary ways) to achieve QSUR scientific goals.

2. Case studies - Concentrate on developing focused case studies using QSURs for various specific problem formulations; this is a key need that can be initiated in the near term.

3. Data collection barriers- Opportunities to collect, curate, and aggregate data while maintaining confidentiality have emerged and should be further explored (an example could be facilitating potential opportunities for sectors to provide frame formulation data for industrial products).

4. Data capture – Further develop data collection processes, consistent data structure and formats for curation, and processes for storage and use of relevant data.

5. Integration - Consider how best to combine and integrate all the identified information, processes, and learnings from case studies.

6. Accelerate communication of QSUR science and applications- Focused outreach activities to exposure scientists, the general exposure and risk assessment regulatory science community, and other stakeholders.

Data availability

No data were used for the research described in the article. The article is a synthesis of QSUR Summit discussions; the full Summit report is included in Supplementary Data.