Nadège Piclin

CAESAR models for developmental toxicity

BackgroundThe new REACH legislation requires assessment of a large number of chemicals in the European market for several endpoints. Developmental toxicity is one of the most difficult endpoints to assess, on account of the complexity, length and costs of experiments. Following the encouragement of QSAR (in silico) methods provided in the REACH itself, the CAESAR project has developed several models.ResultsTwo QSAR models for developmental toxicity have been developed, using different statistical/mathematical methods. Both models performed well. The first makes a classification based on a random forest algorithm, while the second is based on an adaptive fuzzy partition algorithm. The first model has been implemented and inserted into the CAESAR on-line application, which is java-based software that allows everyone to freely use the models.ConclusionsThe CAESAR QSAR models have been developed with the aim to minimize false negatives in order to make them more usable for REACH. The CAESAR on-line application ensures that both industry and regulators can easily access and use the developmental toxicity model (as well as the models for the other four endpoints).

Predicting logP of pesticides using different software

We compared experimental and calculated logP values using a data set of 235 pesticides and experimental values from four different sources: The Pesticide Manual, Hansch Manual, ANPA and KowWin databases. LogP were calculated with four softwares: HyperChem, Pallas, KowWin and TOPKAT. Crossed comparison of the experimental and calculated values proved useful, especially for pesticides. These are harder to study than simpler organic compounds. Structurally they are complex, heterogeneous and similar to drugs from a chemical point of view. They offer an interesting way to verify the goodness of the different methods. Other studies compared several logP predictors using a single set of experimental values taken as a reference. Here we discuss the utility of the different logP predictors, with reference to experimental data found in different databases. This offers three advantages: (1) it avoids bias due to the assumption that one single data set is correct; (2) a given predictor can be developed on the same data set used for evaluation; (3) it takes account of experimental variability and can compare it with the predictors variability. In our study Pallas and KowWin gave the best results for prediction, followed by TOPKAT.

Global QSAR models of skin sensitisers for regulatory purposes.

BackgroundThe new European Regulation on chemical safety, REACH, (Registration, Evaluation, Authorisation and Restriction of CHemical substances), is in the process of being implemented. Many chemicals used in industry require additional testing to comply with the REACH regulations. At the same time EU member states are attempting to reduce the number of animals used in experiments under the 3 Rs policy, (refining, reducing, and replacing the use of animals in laboratory procedures). Computational techniques such as QSAR have the potential to offer an alternative for generating REACH data. The FP6 project CAESAR was aimed at developing QSAR models for 5 key toxicological endpoints of which skin sensitisation was one.ResultsThis paper reports the development of two global QSAR models using two different computational approaches, which contribute to the hybrid model freely available online.ConclusionsThe QSAR models for assessing skin sensitisation have been developed and tested under stringent quality criteria to fulfil the principles laid down by the OECD. The final models, accessible from CAESAR website, offer a robust and reliable method of assessing skin sensitisation for regulatory use.

Chemomentum - UNICORE 6 based infrastructure for complex applications in science and technology

Chemomentum, Grid Services based Environment to enable Innovative Research, is an end-user focused approach to exploit Grid computing for diverse application domains. Building on top of UNICORE 6, we are designing and implementing a flexible, user-friendly Grid system focussing on high-performance processing of complex application workflows and management of data, metadata and knowledge. This paper outlines Chemomentum vision, application scenarios, technical challenges, software architecture and design of the system.

Prediction of oral bioavailability by adaptive fuzzy partitioning

An adaptive fuzzy partition (AFP) algorithm was applied on two bioavailability data sets subdivided into four ranges of activity. A large set of molecular descriptors was tested and the most relevant parameters were selected with help of a procedure based on genetic algorithm concepts and stepwise method. After building several AFP models on a training set, the best ones were able to predict correctly 75% of the validation set compounds. Furthermore, an improvement of about 15% in the validation results was got, on the same data set, as regard to other prediction methods. The importance to work with data sets including a large molecular diversity, and to use tools able to manage it, was also shown. The prediction power was increased up to 25% employing a data set with a better-optimised molecular diversity.

VALIDATION OF THE MODELS

Although many international regulatory bodies recognize the potential benefits of quantitative structure–activity relationship (QSAR) techniques, they are scarcely used in real applications. Some general principles are listed, but the lack of guidelines and standardized protocols accepted and used by all research groups prevents an effective world-wide development of such strategies. The use of appropriate statistical validation tools, such as the training and test set or others, should be adopted for predictive models not only in the case of QSAR based on descriptors but also in the case of models based on rules. In other words, the rules that are defined as appropriate for predictive purposes should also be validated. The statistical tools should prove the capability of the model to be valid in a general way—to be predictive for compounds not used in development of the model.

Chapter 7 – Results of DEMETRA models

The overall process in the context of DEMETRA models involves a careful selection of the data, a check of the chemical structures, and the calculation of thousands of descriptors and fragments, and on that basis a development of hundreds of models. Current computer techniques allow the exploration of a huge space of possibilities in a short time, facilitating the task. This chapter explores a full battery of models. Many of the models are not valid, and the performances are poor. However, a certain number of models give interesting results. Good results are obtained with the use of different models and different chemical descriptors. The heterogeneity of the methodologies increases the robustness of the results, once comparable results are obtained. Indeed, one model can support the other, especially when the starting point and methodology are different.

Classification of a large anticancer data set by Adaptive Fuzzy Partition

AbstractAn Adaptive Fuzzy Partition (AFP) algorithm, derived from Fuzzy Logic concepts, was used to classify an anticancer data set, including about 1300 compounds subdivided into eight mechanisms of action. AFP classification builds relationships between molecular descriptors and bio-activities by dynamically dividing the descriptor hyperspace into a set of fuzzy subspaces. These subspaces are described by simple linguistic rules, from which scores ranging between 0 and 1 can be derived. The latter values define, for each compound, the degrees of membership of the different mechanisms analyzed. A particular attention was devoted to develop structure–activity relations that have a real utility. Then, well-defined and widely accepted protocols were used to validate the models by defining their robustness and prediction ability. More particularly, after selecting the most relevant descriptors with help of a genetic algorithm, a training set of 640 compounds was isolated by a rational procedure based on Self-Organizing Maps. The related AFP model was then validated with help of a validation set and, above all, of cross-validation and Y-randomization procedures. Good validation scores of about 80% were obtained, underlining the robustness of the model. Moreover, the prediction ability was evaluated with 374 test compounds that had not been used to establish the model and 77% of them were predicted correctly.

Use of neighborhood and stratification approaches to speed up instance selection algorithm

This paper investigates a method for instance selection in the context of supervised classification adapted to large databases. Based on the scale up concept, the method reduces the time required to perform the selection procedure by enabling the application of known condensation instance techniques to only small data sets instead of the whole set. The novelty of our approach relies in the way of hybridizing neighborhood and stratification approaches. The key idea is to consider instances found out for a given strata to generate sub populations for the other strata representing critical regions of the feature space. Experiments performed with various data sets revealed the effectiveness and applicability of the proposed approach.