Search

Objectives The current study aims to determine the effect of physicochemical descriptor selection on models of polydimethylsiloxane permeation. Methods A total of 2942 descriptors were calculated for a data set of 77 chemicals. Data were processed to remove redundancy, single values, imbalanced and highly correlated data, yielding 1363 relevant descriptors. For four independent test sets, feature selection methods were applied and modelled via a variety of Machine Learning methods. Key findings Two sets of molecular descriptors which can provide improved predictions, compared to existing models, have been identified. Best permeation predictions were found with Gaussian Process methods. The molecular descriptors describe lipophilicity, partial charge and hydrogen bonding as key determinants of PDMS permeation. Conclusions This study highlights important considerations in the development of relevant models and in the construction and use of the data sets used in such studies, particularly that highly correlated descriptors should be removed from data sets. Predictive models are improved by the methodology adopted in this study, notably the systematic evaluation of descriptors, rather than simply using any and all available descriptors, often based empirically on in vitro experiments. Such findings also have clear relevance to a number of other fields.

This chapter deals with an important aspect of skin permeability which generally sits outside the field of predictive model development—finite-dosing and non-steady-state modelling. This is an area generally but not exclusively associated with the occupational risk exposure and industrial chemical fields. It relates to “real” systems where the controls and conditions described in Chap. 2, and used as the basis for a number of the experiments these models rely on, do not apply. Such systems are difficult and challenging to work with, and this chapter discusses approaches to solving these problems, including the formulation-focused work on finite-dose systems by Riviere which include blends of descriptors, including those which highlight solubility and formulation issues, rather than those which describe the physicochemical properties of the penetrants.

This chapter is a comprehensive critique of algorithms for percutaneous absorption. Using the relationship in a model between its input and output, it critiques the nature, development applicability and use of these models in terms of the wider understanding of their context and appropriate application. While highlighting the benefits of these models—which are mostly related to mechanistic insight and predictive ability—the misuse of models is discussed; this includes how nonlinear models may be prone to over-fitting and adopting redundant descriptors in model development, while also addressing the limitations of linear models, particularly highlighting situations where such models clearly fail to reflect experimental observations. Overall, this chapter seeks to find common ground between these different methods and highlights benefits and pitfalls in their use, which should always be anchored in the experimental situation to which the model is applied and which should always understand how—and from which sources—a model has been developed.

This chapter describes the advances made in methodological aspects of modelling skin permeability by considering artificial intelligence, fuzzy logic and advanced Machine Learning methods and how they have been applied to skin absorption. It reviews the often-stunted contribution to the field of these methods and describes the advantages such methods have produced in recent years. In addition, the chapter also considers problems with these methods, particularly the lack of use in the skin permeability field due to the need for specific computational expertise that is not normally associated with the development of quantitative structure–permeability models, and discusses how such lack of widespread use can be tackled. In addition, as this chapter describes models that often fall into the “black box” category and do not therefore present a readily usable algorithm, an explicit output methods used to discern mechanistic aspects of the skin permeability process are discussed. These include feature selection methods, which have been applied to skin permeability and which have shown the non-independence and interrelationship of key descriptors on each other and that the need for an algorithm with discrete expressions may be of limited value.

This chapter considers those approaches to modelling skin absorption that sit outside the “mainstream” of models based on the approach of Flynn and Potts and Guy and which are often presented as isolated or single diversions into novel aspects of skin permeation. While understanding that the underlying methodology in such studies does not generally change, this chapter describes models developed for non-human species, including rabbit and guinea pig, and how variations of experimental protocols (such as pretreatment and temperature) influence the output of models. It also examines the influence of formulation/vehicles effects on skin absorption, something which is not often considered in other models. Finally, the flux-based models of Roberts are considered in the context of their mechanistic insight and how they can be used to improve the real-world application of algorithms of skin permeability.

This chapter summarises the preceding sections of the book and proposes a series of recommendations for model development and use. These recommendations are based on proposals more widely described in the field of quantitative structure–activity relationships (QSARs) by Cronin, Dearden and others. Implicitly, they suggest that researchers modelling skin permeability should look outside their own field and construct models in a way that is consistent with researchers in other fields. Key findings in the field are discussed, including the significance for data quality of Chilcott’s multicentre validation study and Johnson’s work on the consistency of steroid permeability data, how differing experimental protocols and anomalous data may influence model outputs and the application of predictive software in the generation of physicochemical data. Thus, a series of recommendations on the quality and use of source data, methodological approaches and model transparency are proposed.

Following the pioneering work in this field by El Tayar, Flynn and Potts and Guy—described in Chap. 3—this chapter describes the work that followed and which saw, in the 1990s onwards, a large expansion of interest in this field. This lead to the proliferation of a number of new QSAR models, most of which involved similar methodological approaches and the use of the same data—often as specific subsets focusing on particular classes of chemicals, such as congeneric series or non-electrolytes. In parallel with this, other researchers—notably, Roberts, Hadgraft and Watkinson—proposed through a series of studies an enhanced mechanistic understanding of the skin permeability process and which offers a direct link to the first work in this field through its mechanistic insights when a large number of other studies were simply developing quantitative expressions of limited applicability. Issues of data quality are also discussed in the context of studies by Johnson and Cronin, in particular. This chapter also discusses the implications to these analyses, particularly in terms of experimental design, data set size and consistency, relevance of models to a particular “molecular space” and ultimately, the need for new, consistent sources of data from which to construct models.

Predicting the rate of percutaneous absorption across mammalian and artificial membranes is a complex problem. In previous studies, prediction and accuracy are approached using different machine learning models. Results show that Gaussian processes provided the best result, based on a range of statistical measures. In general the ultimate aim of these machine learning experiments is to try to understand, analyze and predict the percutaneous absorption of drugs across human skin. One way to do this is to select the best set of chemical descriptors and the dataset of synthetic (Polydimethyl siloxane, PDMS) membranes, containing so many descriptors, is considered a suitable dataset to use in this study. Hence, one of the main purposes of the study is to use feature selection methods to select the molecular properties that exert the most important influence on percutaneous absorption across PDMS membranes, in the hope that this will better inform studies on human skin.