New approaches to computational power
The EU-supported CORENET(opens in new window) project is constructing brain-mimicking computing devices that utilise networks of chemical reactions as molecular information processing systems. “We were inspired by the human brain, itself a chemical processor. So we set out to demonstrate that dynamic reaction networks, contained within microfluidic environments, can perform high-level computational tasks,” says consortium member Wilhelm Huck(opens in new window), professor and lead of the CORENET team at Radboud University(opens in new window) in the Netherlands. The work is part of drive to advance machine learning that is being undertaken by EU-funded projects.
Chemical responses interpreted by advanced machine learning
The team implemented reservoir computing on microfluidic chips using chemical reaction networks (CRNs) that convert input feedstock molecules and environmental conditions into a pattern of product molecules. A chemical input layer comprised of a solvent, specific pH and salinity, feedstock molecules and catalysts, is kept at a set temperature and forms a reservoir. The reactions of this reservoir start from simple one- and two-carbon feedstock molecules and expand the chemical space to generate dozens of compounds that are structurally more complex. In doing so, they correlate input conditions with compositional output; meaning that a reaction emerges that can then be analysed by artificial intelligence (AI) and machine learning to extract meaningful information.
The future of computing is now!
A central outcome was the successful development and automation of chemical reservoir computing workflows, where chemical inputs flowing through microfluidic reactors produced non-linear outputs that were analysed and interpreted to solve classification and prediction tasks. Protocols for controlling, monitoring and steering these reaction networks were established, marking a significant step towards scalable, molecule-based computing. A landmark article published in ‘Nature’, titled ‘Chemical reservoir computation in a self-organizing reaction network’(opens in new window), highlighted how formose reactions(opens in new window) and related systems can act as molecular computing substrates. Follow-up work is being finalised in which the generalisation of this concept is demonstrated by a performing reservoir computer with a novel CRN developed. “We were thrilled by how successful the approach was and how our extraordinary group of academic and industry scientists came together to take on this grand challenge of combining systems chemistry, microfluidics, metabolomics and AI,” adds project coordinator and associate professor at the Autonomous University of Madrid(opens in new window) Andres de la Escosura(opens in new window). The CORENET project was funded by the European Innovation Council(opens in new window).
