Use of (Q)SAR for regulatory purposes

It is now possible to obtain (Q)SAR predictions for thousands of chemicals with a press of a key. But to what extent can the authorities use this new information to regulate the many chemicals that we surround ourselves with?

Essentially, there are two ways in which the authorities use (Q)SAR. The predictions can either be used to assess individual substances, or they can be used to reach decisions concerning the relative prioritisation of many different substances.

The REACH legislation places great emphasis on the use of (Q)SAR in connection with the evaluation of chemicals. However, in practice it would be extremely rare for a (Q)SAR prediction to be used as the only basis in an assessment of the safety of a chemical.

Use of (Q)SAR under REACH

The new EU chemicals legislation, REACH, imposes a stricter obligation on the chemical industry in relation to testing and verifying the safety of chemicals that are manufactured within or imported into the EU.

The use of (Q)SARs is a pivotal theme in the legislative text and is for example referred to in Articles 1, 13, 25 and in Annex XI. At a general level, (Q)SAR estimates can be used to indicate whether a substance has a particular dangerous property. According to REACH, (Q)SAR predictions can be used in place of experiments on animals when certain conditions are met.

The industry's use of (Q)SAR in REACH registrations

The REACH Regulation has improved the possibility for industry to increased their use of (Q)SAR methods. This is because the use of (Q)SARs makes it possible to reduce the number of expensive experiments on animals, which can also have ethical implications.

However, the industry and the authorities face a major challenge in ensuring that unreliable (Q)SAR predictions are not used to document the safety of chemicals.

It is therefore important that industry enclose adequate documentation in order tomake it possible to carry out a subsequent independent assessment of the (Q)SAR predictions that form the basis for an assessment of a chemical substance.

This documentation is ensured by following the OECD’s (Q)SAR validation principles and by enclosing documentation of the validity of the model and the prediction (see the section on validation of (Q)SARs on this page).

The authorities use (Q)SAR to prioritise between the many registered substances

REACH obligates the industry to carry out studies, prepare safety assessments of chemicals and submit registrations with this information to the European Chemicals Agency (ECHA). ECHA and the competent authorities of the EU countries (including the Danish Environmental Protection Agency) will select random samples of the industry's chemical evaluations for further examination.

In this work, (Q)SARs are used to prioritise between the many chemicals. The aim is to target the resources in order to assess the chemicals that appear to be the most problematic as regards effects on humans and/or the environment.

(Q)SAR is also used to identify potential problem substances

(Q)SARs are also used to systematically identify potential problem substances, e.g. CMR substances (substances that are carcinogenic, mutagenic or toxic to reproduction) and PBT substances (substances that are persistent, bioaccumulative and toxic).

When the authorities have identified a problem substance on the basis of a (Q)SAR, it may be subject to a more thorough investigation, during which all available information on the substance is taken into consideration. If the results of this investigation indicate that there are grounds for concern, the authorities can propose stricter EU regulation under REACH. They can also request that industry obtains further data through testing.

Read more about REACH on the Danish Environmental Protection Agency website with a link to the legislative text (in Danish) 

The European Chemicals Agency (ECHA) has prepared guidelines on the use of (Q)SAR under REACH. These can be found via the following links:

Guidance on (Q)SARs and grouping of chemicals
( )
Practical guidelines on reporting (Q)SAR, grouping, read-across and weight of evidence
( )

Weight of evidence

In practice, it is relatively rare for a (Q)SAR estimate to be used as the only basis for the authorities' decision concerning the regulation of a chemical. Instead, a so-called weight of evidence (WoE) procedure is often used. With this procedure, all available information is taken into consideration in order to draw an overall conclusion concerning a chemical.

This means that (Q)SAR predictions are used together with available test data, along with the knowledge that is already available concerning other chemicals with a similar structure (grouping and read-across ).

Using WoE, it may be possible to reach a decision in relation to the regulation of a chemical, even if no test data is available from (animal) testing carried out in accordance with applicable guidelines.

Work relating to (Q)SAR within the OECD

The work of developing and validating traditional test methods for chemical substances is carried out under the OECD. The OECD has also undertaken to mediate and promote the use of (Q)SAR methods. Similarly, the OECD will also prepare guidelines for the uniform use and interpretation of (Q)SARs.

The OECD has appointed a working group consisting of experts and authority representatives, within which the use and development of (Q)SARs will be discussed on an ongoing basis. This work has for example resulted in the preparation of a series of guidance documents concerning (Q)SARs. These documents can be found on the OECD website.

The OECD is also currently developing a comprehensive expert system known as the OECD QSAR Application Toolbox. This can be downloaded from the OECD website.

Link to (Q)SAR on the OECD website (,3746,en_2649_34365_33957015_1_1_1_1,00.html ).

Validation of (Q)SAR

When chemicals are studied using traditional test methods, various test guidelines (OECD Test Guidelines) which have been validated internationally are used. Tests carried out in accordance with such guidelines are generally considered to produce reliable and valid results. For (Q)SAR models, the situation is different.

There has been discussion as to whether (Q)SAR models should be validated in the same way as traditional test methods. However, it has been decided that this practice is not appropriate within the (Q)SAR field. This is due to several factors. For example, (Q)SAR models will often be altered on an ongoing basis in order to include new chemicals in the training set or as a result of the adjustment of some of the model's parameters.

It is also difficult to establish ultimate goals for how good a model's predictions should be in order to be accepted, as there can be big variations in the level of precision (or sensitivity or specificity) that can be accepted in relation to the given purpose.

There could for example be a requirement for high precision if the estimate is to be used to classify a chemical. A lower precision could be accepted if the estimate is to be used to prioritise a chemical for further investigation. Finally, the credibility of an estimate depends not only on the model's precision, but also on how convincing the prediction is within or outside the model's area of validity (applicability domain).

The OECD validation principles

Instead of a formal validation of (Q)SAR models, it has been decided that users of a (Q)SAR model have a responsibility to assess the credibility of the (Q)SAR estimates in relation to the given purpose. Model predictions must furthermore be documented both in relation to whether they fall within the (Q)SAR model's area of validity and in relation to how the model itself has been validated.

Provision has therefore been made for the maximum possible transparency in relation to the use of (Q)SAR models. This gives users an opportunity to carry out a qualified evaluation of the model's usability in relation to a given purpose. The OECD has established five validation principles that a valid (Q)SAR model should fulfil:

  1. A defined endpoint (type of effect, e.g. skin irritation)
  2. An unambiguous algorithm (mathematical model which forms the basis for the model)
  3. A defined domain of applicability (area of validity)
  4. Appropriate measures for goodness-of–fit, robustness and predictivity
  5. Be based on a clear chemico-biological mechanism of action

The validity of a (Q)SAR model is documented using what is known as a QMRF ((Q)SAR Model Reporting Format), whilst an assessment can be made as to whether the (Q)SAR predictions fall within the model's area of validity using a QPRF ((Q)SAR Prediction Reporting Format).

Read more about the validation of (Q)SARs on the OECD website: