Application of evolutionary algorithms to optimise one- and two-dimensional gradient chromatographic separations

Method development is an essential procedure to achieve the full potential of a chromatographic separation. However, since it is intrinsically a multi-parameter optimisation problem, this procedure can be quite complex and time-consuming. This is especially true for two-dimensional chromatography, where twice the number of parameters have to be optimised. When applying classical optimisation methods, this leads to a drastic increase in complexity and time consumption.

With the advent of artificial intelligence, novel possibilities to search among numerous combinations have arisen. These include so-called evolutionary algorithms, which mimic biological evolution to solve multi-parameter optimisation problems. They are less sensitive to local optima, as well as to an increase in the number of parameters.

In the present contribution, three evolutionary algorithms were applied to the optimisation of gradient parameters of chromatographic separations: a genetic algorithm, a non-adaptive evolution strategy and a covariance matrix adaptation evolution strategy. In a first phase, they were adapted to method development to reduce their time consumption. In a second phase, their time consumption was compared to that of a grid search method. Both first and second phase were based on in silico meta-experiments, simulating the chromatographic separation of numerous computer-generated samples, to guarantee statistical significance of the results.

The first phase led to insights in the design of evolutionary algorithms. Somewhat surprisingly, the ‘one individual per generation’-design appeared to be the most efficient. The second phase showed that the evolutionary algorithms have a considerable advantage over the grid search method, which increases with the complexity of method development. Notably, this advantage is most pronounced in the case of two-dimensional chromatography, demonstrating the potential of evolutionary algorithms for method development in hyphenated techniques.

Since evolutionary algorithms are versatile, future research could improve their performance by combining them with other optimisation methods or extend their applicability towards other hyphenated techniques.