Balancing properties in Lead Optimisation
Two of the major problems currently impacting productivity in drug discovery are the quality of initial hits and leads against targets, and the time taken to optimise those leads to candidates. Huge amounts of data are generated during the drug discovery process, whether in silico, in vitro or in vivo, and each of these data points will have a degree of error or uncertainty associated with it. Taking account of this uncertainty is essential to making balanced decisions on compound progression. Often it is ignored and compound prioritisation is based on filtering out those which fail to reach a specified value in a particular screen. This approach can result in some of the best balanced compounds being discarded even if they only just fail one criterion. StarDrop's probabilistic scoring algorithm has been designed to accept all available experimental and predicted chemical, biological or ADME data, along with their associated uncertainties, and enable project teams to reach objective decisions on the most appropriate compounds to progress for their specific target.
In vitro potency, selectivity, solubility and microsomal stability data had been generated for a set of 150 client compounds. Compounds had previously been selected for in vivo study based on selectivity and potency, ignoring potential solubility and metabolic stability problems and resulting in poor bioavailability in rats. We were asked to select compounds with a balanced set of properties for progression in vivo.
Profile 1 - Selectivity and potency only
Historically, compounds were filtered and ranked on the basis of their selectivity and potency alone, with selectivity being the most important criterion. This approach did not take into account the errors and uncertainties in the experiments. The table on the right shows the top 15 compounds when ranked by this method. Highlighted is compound XXX572, which was neither the most selective nor the most potent compound in the set. Its relative position in the rankings will be followed throughout this case study.
Profile 2 - Factoring in uncertainty
In consultation with the client, estimates were made of the experimental uncertainties in the assays they were using. These data were imported into StarDrop and the compounds were re-scored. In the table opposite, some compounds, now ranked according to Profile 2, shifted significantly in rank. Compound XXX561 jumped from 28th to 12th position as it was extremely potent and, while it previously "failed" the selectivity cut-off of 8-fold, there was a relatively high probability that its true selectivity was indeed greater than 8-fold due to the uncertainties in the selectivity measurement.
Profile 3 - All available in vitro data
Finally, taking into account all of the in vitro data along with accompanying statistics relating to experimental uncertainties, we saw a considerable change in the compound order. Compound XXX572 was now on top because it satisfied four out of the five criteria. XXX518 came second, as the only compound to satisfy all three of the ADME criteria with potency and selectivity data that, based on assay statistics, were not significantly below the required levels.
Parallel optimisation is essential to developing compounds with the correct balance of properties. In this study, we identified four compounds that had been overlooked by traditional compound selection based on selectivity and potency cut-off values. When tested in vivo, one of these compounds, XXX518, the only synthesised representative of a novel chemotype, was found to have a superior PK profile. Project chemists have now expanded this series, investigating ways of improving selectivity and potency in what appears to be a "Good ADME" chemotype. This new chemistry would not have been considered without the use of StarDrop.
If you'd like more information about the ways StarDrop has been used by our customers, or would like to arrange a demonstration, please contact us. You can also download our product information or try out a free trial version.