According to REACH Guidance (R6), to demonstrate the predictive ability and robustness of the model, a QSAR Model Reporting Format (QMRF) must be provided. This document is well structured and contains all necessary information needed to fulfil the five OECD principles. At the same time, for each test chemical used for prediction purposes, a QSAR Prediction Reporting Format (QPRF) needs to be provided. This report provides all the necessary information to support the value for the predicted endpoint as well as to specify how well it fits within the Applicability Domain for that model.

The applicability domain of the QSAR will be provided to you the moment you request a quote. If KREATiS is not confident that your molecule is covered you will be told before signing a quote. If you wish to verify the applicability domain further to receiving the report, refer to the QSAR Model report  where a complete section is dedicated to identify the chemical groups to which the model can be reliably applied. There are several well-known approaches to define the Applicability Domain of the QSAR models, thus depending upon the algorithm implemented the strategy used to address the reliable chemical space may differ. If your substance of interest satisfies all the conditions specified for the model’s applicability, you can rest assured that the endpoint value has been reliably predicted.

A tutorial will be shortly made available on this website to provide an introduction to the existing Applicability Domain approaches.

As the name suggests, HA-QSARs are aimed to provide High Accuracy QSAR predictions minimising uncertainty or inaccuracy in the predicted endpoint. Moreover, contrarily to the majority of QSARs, the confidence in the result is statistically determined in the same way as it would be for an experimental study. The KREATiS team makes sure that each phase of the modelling strategy from data selection to model validation is carried out in a strict manner using the best quality, validated experimental data. Each model is approved by dedicated QSAR model development and model validation experts at KREATiS. Our aim is to provide results that are at least as good as the best available experimental data. Moreover, we try to ensure that each model is grounded in sound science with mechanistic explanations for the activity relationships observed. Our money back guarantee further reflects upon the quality of QSAR services we offer.

KREATiS follows a strict protocol for model development and validation; therefore each model developed using clients inputs can be tailored according to your prerequisites only after the validity of their datasets has been verified. To better assist with clients’ preferences, the KREATiS consulting team can provide you with an optimal modelling strategy to meet your objectives. The entire procedure is carried out with complete confidentiality.

There is no formal strategy adopted either by the QSAR modellers or the regulatory authorities to define applicability domain of the QSAR models. In fact, there are two sets of applicability domain approaches popular within the QSAR community, first solely based on descriptor values while the second also takes into account the prediction accuracy. For simplicity, they are usually indicated as descriptor-based and property-based approaches.

At KREATiS, we perform several analyses to ensure the predicted endpoint values are reliable, a crucial part of which is to make sure that the requested chemical falls within the defined limits of the model’s reliable predictions. Theoretically, any QSAR model is restrictive, indicating that it could only provide reliable predictions for limited classes of chemicals which were used for the training purposes. If a requested chemical structure happens to fall outside the applicability domain of the iSafeRat models, we will get back to you informing that the predicted values can only be used as approximation as the prediction accuracy was quite low. We will provide you with a probability of the result being accurate. At this point it will be your choice if you wish to use the value perhaps as part of a weight of evidence approach, or move to another option for data gap filling.

Structural similarity and the range of descriptor values are the two parameters commonly used. The former indicates how a chemical is structurally related to the training set. The latter, associated with a query chemical, must be within the limits of the descriptor values defined for the training set.

Not always! If a model was developed using several descriptors (in some cases the number of descriptors actually exceeds the number of training molecules!), it might be at risk of being “over-fitted”. In such cases, the statistical parameters at a first glance may appear outstanding but in reality the model may not be able to accurately predict test molecules not already in the training set. To avoid over-fitting models, cross validation is often performed.

According to the OECD principles, a modeller is advised to provide a mechanistic interpretation of the model, whenever possible. This enhances the transparency in the working principle of a model and allows the regulatory authorities to get better understanding of the modelling approach while providing a basis for its accuracy. However, it is not always a simple matter to correlate the model descriptors to the endpoint, especially if the derivation of the descriptors involved complex formulae or if multiple descriptors are used which prevent relating the result to a scientific theory or law. This is commonly the case for multi-descriptor-based QSARs which are now starting to wane in popularity.

According to ECHA:

PRINCIPLE OF PROPORTIONALITY. The relationship between the amount and quality of data needed and the severity of the decision to be made.

PRINCIPLE OF CAUTION. The relationship between the amount and quality of data needed and the consequence of the decision based on that information being wrong:

-A (Q)SAR result should be “fit for purpose”

-The (Q)SARs perform better on some endpoints than on others

For these reasons KREATiS is compiling collection of validation case studies which will help the authorities to have faith in the results across a wide range of endpoints and highlight the precision of the applicability domains.