The world of chemistry is defined and bounded by the concept of the chemical bond. Chemical reactivity, synthetic accessibility, physical properties and, of particular relevance to the pharmaceutical industry, patentability all rest on this foundation as does the informatics of chemistry; how to organize, store and retrieve chemical knowledge, rests on bond-pattern representations of molecules. However, it is important not to stretch even successful paradigms beyond their domain of consequence. The phenomena of ligand affinity, bioavailability and formulation depend on the variables of molecular interaction. The appropriate variables here are the available shapes and electrostatic profiles of a molecule. These are field-centric properties, not as easily represented as chemical connectivity information. They can, none-the-less, be rigorously defined and manipulated. This talk will describe implementations of this approach, in particular for large-scale applications.

Selection of primary positives in HTS is commonly made on the basis of potency above a particular cut-off of activity that accommodates logistics. Hence, some putatively weak, but still valuable, true hits may have to be abandoned. Moreover, apparently weak compounds may turn out to be potent if they were actually being tested at a concentration lower than that intended. The reliability of the activity value highly depends upon the quality of the assay and the distribution of activity for the population of samples tested. In all, a hit selection process minimizing rates of false positives and negatives would optimize the usage of limited screening resources. We will comment on statistical methodologies that can be efficiently applied to improve the probability of success of the hit selection process in HTS. These include usage of a statistical cut-off based on the variability of the population of inactive samples from the HTS campaign.

In earlier work done in our laboratory, we had proposed the use of the ring-system and the ring-scaffold to cluster structurally related compounds. The use of ring-systems and ring-scaffolds allows for rapid compound clustering, novelty analysis, and compound acquisition. We had also described the use of the ring-scaffold for the design of undirected combinatorial screening libraries.

Here we extend the concept to a whole family of structural fragments. These structural fragments include: whole molecule, ring-scaffold, topological ring-scaffold, ring-system pair scaffold, ring-system pair, bridge, exo-scaffold substituent, ring-system collapsed scaffold, and functional group. We have created an Oracle database to permit facile retrieval of these fragments.

We will discuss these fragments and show how they can be used to analyze HTS data, discover structure-activity relationships, and to mine structural databases in novel ways.

Traditional HTS is a random process with the aim of screening as many structurally distinct compounds as possible. A more efficient approach is to apply focused iterative screening (FIS) for instances where targets do not permit an HTS campaign. The rational is to use a small subset of postulated actives enriched with inactives and a powerful statistical modeling tool, Random Forest, to select a biased test set of molecules from a large compound library. This results in higher quality data by testing samples at 3 doses followed by confirmation of actives before the next round of FIS. Our method rapidly identifies true hits and is more successful than others to reduce the number of screening iterations and test samples. Results from two examples of FIS will be shown and compared with real HTS data to demonstrate our approach increases the hit rate and is successful in identifying novel lead structures.