OECD QSAR Toolbox Uses
What is the OECD QSAR Toolbox used for?
The OECD QSAR Toolbox is free software for chemical hazard assessment. You use it to fill gaps in (eco)toxicity data without new testing: it groups a chemical with structurally or mechanistically similar substances and then predicts the missing endpoint by read-across, trend analysis or QSAR models. The Toolbox website says it supports "reproducible and transparent chemical hazard assessment" (qsartoolbox.org).
What it does (qsartoolbox.org):
- Data retrieval: it searches about 155,000 chemicals across 63 databases, which hold more than 3.3 million experimental data points.
- Profiling: it checks a chemical against structural alerts and mechanistic profilers to find its structural and mechanistic properties.
- Analogue and category building: it finds structural or mechanistic analogues, builds chemical categories and checks whether each category is consistent.
- Metabolism simulation: it predicts metabolites and degradation products, so these can also be considered in grouping.
- Filling data gaps: it predicts a missing endpoint by read-across, trend analysis or external QSAR models built into the Toolbox.
- Reporting: it produces data matrices and prediction reports that you can customise for documentation.
How read-across works in it. The OECD tutorial on skin sensitisation (Toolbox v.4.4.1) explains the analogue approach. Endpoint data for one or a few tested chemicals are used to predict the same endpoint for a "similar" untested chemical. The approach is "especially appropriate for 'qualitative' endpoints" such as positive, negative or equivocal results. It rests on the hypothesis that the members of a category behave in a common way toxicologically (OECD QSAR Toolbox tutorial, p. 5).
Regulatory context. The OECD built the Toolbox together with ECHA. It is free, and the current version is v.4.9. Regulatory agencies, industry and research institutions use it for chemical safety assessment under frameworks such as REACH and pesticide regulation (qsartoolbox.org). OECD also publishes Toolbox tutorials on ECHA's Read-Across Assessment Framework (RAAF) scenarios. That is the framework ECHA uses to evaluate read-across arguments in REACH dossiers (RAAF Scenario 1 tutorial, RAAF Scenario 4 tutorial).
Approach: I searched the regulatory guidance I hold that covers QSAR and read-across: the SCCS Notes of Guidance, Japan NITE's QSAR and HESS pages, and the Danish EPA's QSAR-based advisory list. None of those searches finished, so I relied on the official QSAR Toolbox website and OECD's Toolbox tutorials. I couldn't open OECD's own Toolbox page (oecd.org), which refused access.
Can I use its predictions for REACH?
Yes, Toolbox predictions can go into a REACH dossier, but only as a justified adaptation of the standard testing requirements, not as an automatic replacement for a study. ECHA, not the OECD, decides whether a Toolbox result is accepted under REACH. The Toolbox website sends questions "concerning policy or regulatory acceptance of Toolbox results within the EU legislations of ECHA's competence" to ECHA itself (qsartoolbox.org).
The legal basis. The conditions for using (Q)SAR results, and for grouping and read-across, are set in REACH Annex XI (Regulation (EC) No 1907/2006). I couldn't open the REACH text this time, either in the EU legislation datasets or on EUR-Lex. So I can't quote the conditions here, and I'm not paraphrasing them from memory. ECHA's practical guide "How to use and report (Q)SARs" covers the same ground, but I couldn't open it either: ECHA's site refused access (ECHA practical guide).
How an assessor will judge a prediction. The OECD (Q)SAR Assessment Framework (QAF, 2nd edition, 2024; OECD Series on Testing and Assessment No. 405) was co-led by ECHA. It is the main reference for how regulators review (Q)SAR results:
- A valid model is not enough. The framework says the assessment "should not be limited to checking the validity of the model used because even a valid model can produce unacceptable predictions in certain conditions." Each prediction needs its own assessment (QAF, p. 7).
- The model is checked against the five OECD validation principles. These are a defined endpoint, an unambiguous algorithm, a defined applicability domain, appropriate measures of goodness-of-fit, robustness and predictivity, and a mechanistic interpretation, if possible (QAF, p. 11).
- Each prediction is checked against four principles: the input was correct, the substance falls within the model's applicability domain, the prediction is reliable, and the outcome is fit for the regulatory purpose (QAF, p. 7).
- What you document. The model developer reports the model in a QMRF. You, as the (Q)SAR user, report the prediction in a QPRF, plus a QRRF if your result combines several predictions. The framework says you "could" fill in the assessors' checklists in advance to make their review easier (QAF, pp. 8-9).
In practice, for the dossier:
- Read-across: build your argument with ECHA's Read-Across Assessment Framework (RAAF) in mind. The OECD Toolbox tutorials on the RAAF scenarios, cited last time, show how (RAAF Scenario 1 tutorial).
- Reports: the Toolbox's customisable prediction reports and data matrices can serve as the documentation that goes with the prediction (qsartoolbox.org).
- What the Toolbox doesn't do: it doesn't settle whether the prediction is good enough for your endpoint or for classification. You have to justify that.
Approach: I tried the REACH legal text and the ECHA Board of Appeal decisions on QSAR and read-across in the datasets, but neither search finished. So I used the OECD QAF and the Toolbox website. EUR-Lex and ECHA's site (including its (Q)SAR practical guide) wouldn't open, so the Annex XI wording is not quoted here.
How does Japan's HESS compare?
HESS is a much narrower tool than the OECD QSAR Toolbox. It covers one endpoint, repeated dose toxicity, and predicts it for untested chemicals using the category approach. The Toolbox covers many (eco)toxicity endpoints. The two are complementary: NITE states that "HESS is compatible with the OECD QSAR Toolbox" (NITE HESS page).
| Japan's HESS (NITE) | OECD QSAR Toolbox | |
|---|---|---|
| Developer | NITE's Chemical Management Center. It began as a project under NEDO and METI (FY2007-2012) involving industry, government and academia. NITE now updates the data, releases the system and shares application examples (NITE 2016, p. 13) | OECD, with ECHA (qsartoolbox.org) |
| Endpoint scope | Repeated dose toxicity only (NITE HESS page) | Many human-health and environmental endpoints |
| Method | Groups chemicals "on the basis of structural, physicochemical and mechanistic similarities" and predicts repeated dose toxicity by the category approach (NITE HESS page) | Profiling, analogue and category building, then read-across, trend analysis or built-in QSAR models |
| Data | HESS DB holds two databases: (1) repeated dose toxicity and toxicity mechanisms; (2) rat metabolism maps and ADME data for rats and humans (NITE HESS page) | About 155,000 chemicals across 63 databases, with more than 3.3 million data points |
| What sets it apart | When released in 2012 it was the world's first prediction system that let users look up toxicity test, metabolism and mode-of-action information for repeated dose toxicity in one place (NITE 2016, p. 13) | Breadth, metabolism simulation, and reports you can customise |
| Regulatory anchor | NITE says regulatory use is "expected", "focusing especially on the risk assessment of those chemicals listed in the Japanese Chemical Substance Control Law" (CSCL) (NITE HESS page) | Used under REACH and other frameworks; ECHA decides whether results are accepted in the EU |
| Access | Free, but you must register; runs on Windows (NITE HESS page) | Free download |
| Recent updates | Repeated dose toxicity data added for 110 substances (March 2022), 56 (March 2023) and 67 (March 2024). Liver and kidney toxicity categories were developed with Kao Corporation (2022). The latest release (July 2026) updated the operating environment (NITE update history) | v.4.9 is current |
The Toolbox column comes from the Toolbox website I cited in my earlier answers (qsartoolbox.org).
What this means for regulatory use:
- Japan has been more cautious on in-silico human-health data. A 2016 NITE-hosted expert commission described the position under the CSCL at that time. QSAR (EPI Suite) and read-across were accepted to fill gaps in physico-chemical properties, biodegradability and bioconcentration for existing-substance screening and risk assessment. For human-health hazard data, the position was that data "should be obtained through testing", and how far in-silico methods could be used was still to be explored (NITE 2016, pp. 7-8). That paper is a discussion document, not a regulatory rule.
- So under the CSCL, HESS works more as a tool for supporting expert judgement than as an accepted way to replace a study. The same commission noted that, at the time, there was no budget for improving HESS beyond maintenance (NITE 2016, p. 25).
- For a REACH dossier, a HESS category for repeated dose toxicity would face the same RAAF and Annex XI scrutiny as any other read-across. Because HESS gives detail on mechanism and metabolism, it can strengthen the similarity argument that RAAF assessors look for. Using HESS doesn't change any of the REACH acceptance conditions discussed earlier.
- In practice, the two tools are often used together: the Toolbox for broad profiling and for endpoints other than repeated dose toxicity, and HESS for mechanistic categories for repeated dose toxicity.
Approach: I tried the NITE QSAR/HESS documentation in the datasets, but the search didn't finish, so I relied on NITE's official HESS pages and its 2016 report on in-silico methods. I found nothing on a current METI/MHLW/MOE position on accepting HESS predictions under the CSCL newer than that 2016 report.