Accelerated insertion of nanocomposites into advanced applications is predicated on the ability to perform a priori property predictions on the resulting materials. In this paper, a paradigm for the virtual design of spherical nanoparticle-filled polymers is demonstrated. A key component of this “Materials Genomics” approach is the development and use of Materials Quantitative Structure-Property Relationship (MQSPR) models trained on atomic-level features of nanofiller and polymer constituents and used to predict the polar and dispersive components of their surface energies. Surface energy differences are then correlated with the nanofiller dispersion morphology and filler/matrix interface properties and integrated into a numerical analysis approach that allows the prediction of thermomechanical properties of the spherical nanofilled polymer composites. Systematic experimental studies of silica nanoparticles modified with three different surface chemistries in polystyrene (PS), poly(methyl methacrylate) (PMMA), poly(ethyl methacrylate) (PEMA) and poly(2-vinyl pyridine) (P2VP) are used to validate the models. While demonstrated here as effective for the prediction of meso-scale morphologies and macro-scale properties under quasi-equilibrium processing conditions, the protocol has far ranging implications for Virtual Design.

For well over 100 years, chemists have explored the relationship between the chemical structure and biological activity, and dreamed of predicting them as well as other measurable properties. The first description of a relationship between composition and activity [1] was based on observations of correlation between specific molecular features and observable physiochemical properties [2]. With some data tabulation, it was found that structure-activity relationships could be used to quantify chemical intuition: For a small change in the molecular structure, a corresponding small change in activity could be explained by analyzing regular changes the numerical representations of molecular structure. The power inherent in this type of relationship quickly became obvious, and increased in importance with the quick tabulation abilities of computers. The reductionist qualities of quantitative structureactivity relationships (QSARs) have resulted in both praise and condemnation for the discipline throughout its existence [3][4][5]. Without debating the philosophical validity of reductionist views, a more practical approach is to understand how and when QSARs are applicable to relevant problems. As discussed below, there are many choices to make when matching available data with types of chemical descriptors and machine learning methodologies (Figure 2.1). Inherent in these choices are decisions that affect the level of difficulty and computational effort needed to develop a model and to establish its domain of applicability -a crucial element for managing end-user expectations of model performance. Most models are constructed using methods that project or compress information into a simpler form, consequently representing a compromise between mode interpretability and predictive power. For any nontrivial QSAR the importance of good chemical descriptors cannot be overstated -even the most capable machine learning methodology cannot extract signal from descriptor variance that is not monotonically related to the endpoint of interest. This is the essential Tao of building QSARs, where the ultimate goal is to construct chemically meaningful, validated models. Achievement of this goal relies Statistical Modelling of Molecular Descriptors in QSAR/QSPR. First Edition. Edited

No abstract

No abstract

A quantitative structure-activity relationship was developed to predict the efficacy of carbon adsorption as a control technology for endocrine-disrupting compounds, pharmaceuticals, and components of personal care products, as a tool for water quality professionals to protect public health. Here, we expand previous work to investigate a broad spectrum of molecular descriptors including subdivided surface areas, adjacency and distance matrix descriptors, electrostatic partial charges, potential energy descriptors, conformation-dependent charge descriptors, and Transferable Atom Equivalent (TAE) descriptors that characterize the regional electronic properties of molecules. We compare the efficacy of linear (Partial Least Squares) and non-linear (Support Vector Machine) machine learning methods to describe a broad chemical space and produce a user-friendly model. We employ cross-validation, y-scrambling, and external validation for quality control. The recommended Support Vector Machine model trained on 95 compounds having 23 descriptors offered a good balance between good performance statistics, low error, and low probability of over-fitting while describing a wide range of chemical features. The cross-validated model using a log-uptake (qe) response calculated at an aqueous equilibrium concentration (Ce) of 1 μM described the training dataset with an r(2) of 0.932, had a cross-validated r(2) of 0.833, and an average residual of 0.14 log units.