The main goal of CORENET is to implement reservoir computing based on reaction networks as an enabling technology to accelerate discovery in the chemical sciences. The implementation of chemical RC will be based on the compartmentalisation of reaction networks with increasing complexity in microfluidic flow reactors.
The output of these reservoir computing systems can be monitored using mass spectrometry or other analytical methods, such as gas chromatography (GC) or liquid chromatography (LC) coupled to mass spectrometry (MS) — i.e. GC-MS and LC-MS, respectively.
A second level of device complexity will be achieved through the coupling of multiple fluidic reactors, al-lowing the generation of a network of chemical RC devices. This will be achieved through 3 interrelated scientific and technological objectives.
Interfacing reaction networks with metabolomics mass spectrometry and cheminformatic tools for their screening and control
The aim is to explore reaction patterns (e.g., feedback loops and autocatalytic subsets/cycles) in complex reaction networks and establish parallels with the mechanisms of metabolic regulation, thermal homeostasis and biological oscillations.
A new computational methodology for analysing CRNs, which uniquely combines graph theory and algorithms with methods such as machine learning, will be contrasted with large datasets obtained from HPLC-MS or GC-MS analyses.
On-chip integration of complex reaction networks and analytical methods
The aim is to integrate the recurrent reaction networks developed previously into a microfluidic flow system that provides full control over input variables and interfaces with analytical equipment to measure the output of such networks.
More specifically, we will develop flow reactors with (coupled) temperature-controlled continuously stirred tank reactors (CSTRs) or fluidic equivalents thereof.
A second part of this objective will be to develop suitable interfaces with GC-MS and LC-MS analytical methods.
Demonstration of chemical reservoir computing
The aim is to deploy the CRNs developed together with their automated on-chip operation and product analytics for computing purposes.
More specifically, we will develop mapping methods between the CRN’s input layer and product output data for label generation and will use cloud-based tools for the training, deployment and evaluation of machine-learning (ML) models to interpret input–output correlations of various CRNs.
The ML methods will be directed to two possible uses: maximisation of the production rate of CRNs for specific product formations and image recognition by a chemical reservoir computing system.