TR 074 : QSARs in the Assessment of the Environmental Fate and Effects of Chemicals | June 1998
Quantitative Structure Activity Relationships (QSARs) are based on a comparison of the structure of some physico-chemical property of a substance (‘descriptor’) with a measured endpoint which may be another physico-chemical property of a biological effect. The issues that arise in the development, selection and use of QSARs are discussed together with appropriate examples.
The report describes how all QSARs should be based on a defined and measurable endpoint. A consistence dataset for that endpoint covering a well defined scope of chemical structures (‘domain’) is required from which a ‘training set’ to be used for the development of the mathematical correlation is separated. The remainder of the dataset is the ‘test set’ which is used to verify the mathematical correlation. Finally, a clearly described statistical process must be employed in order to determine the variability.
The principal lessons learnt are that the variability inherent in the measured endpoints and descriptors determines the variability in the prediction. How this variability will need to be addressed will depend on the use of the prediction.
QSARs that are developed with an understanding of the relationship between the endpoint and the descriptors, i.e. ‘mechanistic’ QSARs, have a number of advantages over those based only on a statistical relationship between descriptors and the endpoint. These advantages may include being able to better understand how to investigate outliers, whether new chemicals are part of the domain of the original QSAR and if not, being able better to assess the potential for extrapolation.
From the above points it is clear that QSARs should only be used by experts, who need to understand more than the endpoint being predicted. They should also understand the descriptors used and their relevance to the chemicals in the QSAR’s domain, and the use to which the QSAR is to be put. When used properly, QSARs are capable of highlighting new understandings and if the variability is accounted for, may help in the development of probabilistic risk assessments.
However, when QSARs are applied to substances outside of the domain for which they were developed, the uncertainties involved will often lead to the incorporation of added conservatism and increased error propagation within risk assessment models.
From the reviews carried out in this report there are a number of QSARs that need to be developed. These are:
Metabolism in fish;
Bioconcentration including metabolism in organisms other than fish;
Microbial breakdown of chemicals;
Soil/water partitioning including kinetics;
Effects on terrestrial organisms, sediment dwelling organisms and marine organisms.
One major area within the present EU risk assessment process which is poorly served by QSARs is the understanding of how chemicals behave within the terrestrial and sedimentary compartments and to what extent they are capable of expressing their intrinsic properties. Not only is there insufficient information for assessing these compartments, but as a result, there are few QSARs to predict the influence of bioavailability on toxicity.
It is important that QSARs that are recommended for use in the regulatory area should be subject to constant improvement and refinement as more data become available. This is the case within the US – PMN1 process, although it is also recommended that the improvement should be as transparent as possible, given commercial confidentiality.
If is the recommendation of this report that QSARs could be used to check the validity of data or to fill data gaps, for priority setting, risk assessment and classification. However, in such circumstances, there need to be appropriate mechanisms to allow for the generation of measured data when requested. Valid measured data, when available, should always take precedence over QSAR predictions.
QSARs may be defined as models, involving multi-stages, within the endpoint. Thus, a fish toxicity QSAR includes uptake, distribution, metabolism and excretion. As such it is important that they are developed, selected and used by experts with a clear understanding of the uncertainty involved in their prediction and a good understanding of how to take this uncertainty into account, whether for risk assessment of classification.
1 Pre-Manufacture Notice