Autonomous Experimentation: Coupling Active Learning with Computer Controlled Microfluidics (abstract)

The interactions of biomolecular substrates, such as networks of enzymes, exhibit behaviours that could allow for new modes of information processing. Information processing is not required to work within physiological conditions, therefore the contribution of existing knowledge about enzyme interactions is limited. New knowledge about the interactions of enzymes can only be obtained experimentally. However, the high dimensionality of the parameter spaces mean that the resources available are limited compared to the size of the space to explore. Often this can leave at most a handful of experiments per parameter dimension. Additionally, the validity of experimentally obtained observations are not guaranteed, particularly in the biological domain where experimental error can produce observations not representative of the true behaviour. By combining active learning techniques with an automated lab-on-chip platform, we are working towards a fully autonomous machine. This machine will provide effective resource usage, achieved through both minimising the number of experiments required to be performed and by reducing the chemical resources consumed in each experiment