The Surface Site Interaction Point (SSIP) model describes the non-covalent interaction properties of molecules by abstracting a discrete number of points on the van der Waals surface of molecules. Each point is assigned a value based on empirical interaction scales and the calculated Molecular Electrostatic Potential (MEP). SSIPs have been used to provide predictions of partition coefficients, solvent effects on association constants for formation of intermolecular complexes, and the probability of cocrystal formation.

In this thesis, secondary electrostatic interactions are shown to be highly overestimated at the van der Waals surface and to be more accurately described on electron density isosurfaces that lie closer to the nuclei. Interaction parameters calculated with these isosurfaces successfully account for the properties of arrays of multiple H-bond donor and acceptor groups in different configurations. Three MEP isosurfaces are required to describe soft H-bond acceptors, hard H-bond acceptors, and H-bond donors. The Atomic Surface Site Interaction Point (AIP) model has been developed to obtain interaction points on these three surfaces using empirical rules for different atom types. This new approach to obtain interaction sites ensures a correct description of secondary electrostatic interactions, an accurate placement of lone pairs, a consistent description of π-systems and a representation of short H-bond contacts with hard acceptors but longer contacts with soft acceptors.

Partition data between n-hexadecane and water was used to fine-tune the AIP representation of non-polar functional groups, for which it is difficult to obtain accurate empirical interaction parameters. The phase transfer data is also used to analyse the effect of H-bond cooperativity on the AIPs of functional groups that can make more than one H-bond, like alcohols, ethers and carbonyls.

Finally, two methods for fast calculation of AIPs are discussed. The first method is a fragment-based approach, which assigns AIPs of large compounds from the AIP representation of small molecules with matching substructures. The second approach relies on a neural network method developed by Astex to quickly calculate the MEP surface and obtain the AIPs. These methods extend the scope of the AIP model to describing large molecules or large libraries of compounds for applications such as virtual screening and modelling host-guest systems.