Chromatographic analysis represents a large proportion of the analytical testing used to support pharmaceutical drug development as well as quality control.  Even though some recent trends in pharmaceutical analysis seem to favour Process Analytical Technology (PAT) and spectroscopic tools, chromatography remains the number one choice mainly for its ability to deal with complex matrices.  Chromatographic method lifecycle typically consists of three steps, method development, method validation and method application.  This process is continuously applied in parallel with product development as the changes in synthetic process and formulation usually trigger the need for method development or optimisation.

Chromatographic method development is heavily reliant upon laboratory experimentation.  A number of stationary phases are screened using low, medium and high pH mobile phases and generic gradient profiles.  Stationary phases included in these screens are usually selected based on personal preferences or past success.  The stationary phase which provides the best initial resolution of compounds of interest is then taken through a series of experiments during which the pH, gradient and sometimes temperature are optimised, typically using ‘non-systematic’, ‘one parameter at a time’ approach.  This method development process usually ends when a set of conditions, which meet the resolution criteria is obtained.  Although such an approach is often ultimately successful, it generates a huge amount of waste in the form of analytical runs which fail to provide the required resolution. 

The objective of this work is to establish whether it is possible to build sufficiently accurate mathematical retention models based on retention data obtained at fixed experimental conditions.  Molecular descriptors and physico chemical properties of approximately 100 compounds are used to develop structure retention relationships.  We employ genetic algorithm to carry out data reduction and to identify significant descriptors in order to build sufficiently accurate models.  We also discuss the impact of structural similarity on accuracy of these models. 

It is our intention to implement these in-silico models in our method development workflow, to reduce or eliminate the initial screening step and direct the method development activities to a chemical space where the probability of success is high