<p><br></p><p>The pharmaceutical industry depends on deeply understanding pharmaceutical excipients and active ingredients. The physicochemical properties must be sufficiently understood to create a safe and efficacious drug product. High-throughput methods have reduced the time and material required to measure many properties appropriately. However, some are more difficult to evaluate. One such property is solubility or the equilibrium dissolvable content of the material. Solubility is an essential factor in determining the bioavailability of an active ingredient and, therefore, directly impacts the effectiveness and marketability of the drug product.</p><p>Solubility can be a challenging, time-consuming, material-intensive property to measure correctly. Due to the challenge associated with determining experimental values, researchers have devoted a significant amount of time toward the accurate prediction of solubility values of drug-like compounds. This remains a difficult task as there are two hurdles to overcome: data quality and specificity of molecular descriptors. Large databases of reliable solubility values have become more readily available in recent years, lowering the first barrier to more accurate solubility predictions. The second hurdle has proven more challenging to overcome. Advances in artificial intelligence (AI) have provided opportunities for improvement in estimations. Expressly, the subsets of machine learning and neural networks have provided the ability to evaluate vast quantities of data with relative ease. The remaining barrier arises from appropriately selecting AI techniques with descriptors that accurately describe relevant features. Although many attempts have been made, no single set of descriptors with either data-driven approaches or <i>ab initio</i> methods has accurately predicted solubility.</p><p>The research within this dissertation focuses on an attempt to lower the second barrier to solubility prediction by starting with molecular features that are most important to solubility. By deriving molecular descriptors from the electronic properties on the surface of molecules, we obtain precise descriptions of the strength and locality of intermolecular interactions, critical factors in the extent of solubility. The novel molecular descriptors are readily integrated into a Deep-sets based Graph and Self-Attention Neural Network, which evaluates predictive performance. The findings of this research indicate significant improvement in predicting intrinsic solubility over other literature-reported methods.</p>

This thesis covers a range of methodologies to provide an account of the current (2010-2014) state of the art and to develop new methods for solubility prediction. We focus on predictions of intrinsic aqueous solubility, as this is a measure commonly used in many important industries including the pharmaceutical and agrochemical industries. These industries require fast and accurate methods, two objectives which are rarely complementary. We apply machine learning in chapters 4 and 5 suggesting methodologies to meet these objectives. In chapter 4 we look to combine machine learning, cheminformatics and chemical theory. Whilst in chapter 5 we look to predict related properties to solubility and apply them to a previously derived empirical equation. We also look at ab initio (from first principles) methods of solubility prediction. This is shown in chapter 3. In this chapter we present a proof of concept work that shows intrinsic aqueous solubility predictions, of sufficient accuracy to be used in industry, are now possible from theoretical chemistry using a small but diverse dataset. Chapter 6 provides a summary of our most recent research. We have begun to investigate predictions of sublimation thermodynamics. We apply quantum chemical, lattice minimisation and machine learning techniques in this chapter. In summary, this body of work concludes that currently, QSPR/QSAR methods remain the current state of the art for solubility prediction, although it is becoming possible for purely theoretical methods to achieve useful predictions of solubility. Theoretical chemistry can offer little useful additional input to informatics models for solubility predictions. However, theoretical chemistry will be crucial for enriching our understanding of the solvation process, and can have a beneficial impact when applied to informatics predictions of properties related to solubility.

<p>Aqueous solubility is a key parameter influencing the bioavailability of drugs and drug candidates. In this thesis computational models for the prediction of aqueous drug solubility were explored. High quality experimental solubility data for drugs were generated using a standardised protocol and models were developed using multivariate data analysis tools and calculated molecular descriptors. In addition, structural features associated with either solid-state limited or solvation limited solubility of drugs were identified.</p><p>Solvation, as represented by the octanol-water partition coefficient (log<i>P</i>), was found to be the dominant factor limiting the solubility of drugs, with solid-state properties being the second most important limiting factor.</p><p>The relationship between the chemical structure of drugs and the strength of their crystal lattice was studied for a dataset displaying log<i>P</i>-independent solubility. Large, rigid and flat molecules with an extended ring-structure and a large number of conjugated π-bonds were found to be more likely to have their solubility limited by a strong crystal lattice than were small, spherically shaped molecules with flexible side-chains.</p><p>Finally, the relationship between chemical structure and drug solvation was studied using computer simulated values of the free energy of hydration. Drugs exhibiting poor hydration were found to be large and flexible, to have low polarisability and few hydrogen bond acceptors and donors.</p><p>The relationship between the structural features of drugs and their aqueous solubility discussed in this thesis provide new rules-of-thumb that could guide decision-making in early drug discovery.</p>

Aqueous solubility is a key parameter influencing the bioavailability of drugs and drug candidates. In this thesis computational models for the prediction of aqueous drug solubility were explored. High quality experimental solubility data for drugs were generated using a standardised protocol and models were developed using multivariate data analysis tools and calculated molecular descriptors. In addition, structural features associated with either solid-state limited or solvation limited solubility of drugs were identified. Solvation, as represented by the octanol-water partition coefficient (logP), was found to be the dominant factor limiting the solubility of drugs, with solid-state properties being the second most important limiting factor. The relationship between the chemical structure of drugs and the strength of their crystal lattice was studied for a dataset displaying logP-independent solubility. Large, rigid and flat molecules with an extended ring-structure and a large number of conjugated π-bonds were found to be more likely to have their solubility limited by a strong crystal lattice than were small, spherically shaped molecules with flexible side-chains. Finally, the relationship between chemical structure and drug solvation was studied using computer simulated values of the free energy of hydration. Drugs exhibiting poor hydration were found to be large and flexible, to have low polarisability and few hydrogen bond acceptors and donors. The relationship between the structural features of drugs and their aqueous solubility discussed in this thesis provide new rules-of-thumb that could guide decision-making in early drug discovery.