Mining the Knowledgebase
Accurate prediction of chiral selectivities based on chemical structures is challenging, as small changes in a single functional group can result in dramatic changes in the effectiveness of a method. However, it has been demonstrated that structure similarity will correlate to a large degree with selectivity. Screens targeted according to structure can thus reduce the number of injections that are required, provided that an effective ranking system can be designed.
Structure Similarity Search. A number of methods have been designed to quantitatively rank chemical structures in terms of similarity, including Tanimoto, Dice, Cosine, and others. While the precise value for the similarity is dependent on the algorithm used, the ranking order is generally the same. This implies that any algorithm incorporating structure similarity terms can be used, but that for consistent evaluation of similarity from day-to-day, the given algorithm should be standardized.
Physicochemical Parameters. Incorporating predicted physicochemical parameters such as polar surface area can, for given methods, provide additional clues as to the chances of success of the method. To develop a more targeted “chiral similarity search”, it is necessary to check the correlation of an array of parameters to chiral selectivity rather than individual parameters that may change with the given method. Obviously, then, it is necessary to compile a knowledgebase for each candidate method before implementing a search algorithm.
For this reason, the most common approach is to specifically target structure similarity search. However, as a knowledgebase grows, the method-ranking system can be enhanced by incorporating physicochemical parameters.
Figure 3 gives a general approach to the creation of a screening strategy based on the results of a combined structure similarity and predicted physicochemical ranking of chromatographic methods. The organization can create customized strategies like these, or use generic approaches based simply on structure similarity. Early work by one laboratory indicated that prioritized screening based on a top six structure similarity approach can closely match the success rate of a full 24-method screen.
In more than 90% of cases, a viable scale-up system was identified using six injections. More than half of the remaining compounds were found to be non-viable using any of the 18 additional methods.
In theory, this kind of system can be used to quantitatively model chiral selectivities that can be expected for each method for a given set of enantiomers. However, this level of rigor is unnecessary. A simple ranking of methods based on the approach, and connection of the results to the automated screening system, is extremely effective.
It is typically reported that as many as 95% of compounds generated by medicinal chemists can be handled with a set of screens designed to address a large variety of compounds—usually less than 50 standard methods. The remaining compounds can provide purification labs with considerable headaches. Applying a structure-searchable knowledgebase, such as ChirBase for ChromManager, can provide over 150,000 candidate methods for the most challenging of compounds.
Modern drug discovery organizations typically generate sufficient data to enable dramatically improved chiral screening throughput. Automatic mining of the various software systems that contain this data can create an appropriate knowledgebase. Structure-based queries can then use this knowledgebase to prioritize screening experiments, reducing the number of injections required to identify viable methods for scale-up. The result is improved sample turnaround and reduced solvent and other instrumentation costs.