Periodic Reporting for period 2 - CORENET (Complex chemical reaction networks for breakthrough scalable reservoir computing)
Berichtszeitraum: 2023-04-01 bis 2024-09-30
Chemical synthesis has traditionally been a discipline that was part clever application of accumulated knowledge and part “art”. In recent years, a revolution has been gathering pace with the introduction of robotics, microfluidics, integrated analytics, and AI-based integration of data, leading to programmable synthesis of molecules using computer-controlled devices. This paradigm change opens up new opportunities to push chemistry beyond the synthesis of molecules to the synthesis of chemical reaction networks (CRNs), which in living organisms control all essential processes.
CORENET aims at construct chemical systems that can perform high-level computation based on chemical reactions. The project will implement reservoir computing based on the compartmentalisation of reaction networks with increasing complexity in microfluidic flow reactors.
The main advantage of CRNs for computing is that they are able to generate vast compositional chemical spaces, allowing them to process information and regulate their responses in a ‘metabolic’ way. The project will bring new knowledge bridging system chemistry, microfluidics technology and computational science and may revolutionize patient treatment via in situ synthesis of drug molecules.
CORENET aims at construct chemical systems that can perform high-level computation based on chemical reactions. The project will implement reservoir computing based on the compartmentalisation of reaction networks with increasing complexity in microfluidic flow reactors.
The main advantage of CRNs for computing is that they are able to generate vast compositional chemical spaces, allowing them to process information and regulate their responses in a ‘metabolic’ way. The project will bring new knowledge bridging system chemistry, microfluidics technology and computational science and may revolutionize patient treatment via in situ synthesis of drug molecules.
During this period, the CORENET consortium has shown that the formose and glyconitrile CRN compositions are highly dynamic and non-linearly dependent on different variable conditions. Additional chemical inputs are currently being explored for both cases, with promising results. Hence, these CRNs can in principle be used to perform advanced computing tasks. We have actually widen the available tools for massive and on-line output data extraction from GC-MS and tims-TOF to techniques such as FT-IR and UV-Vis spectroscopy, which are fast, accurate and cheap. This enables to monitor the chemical evolution and the steering properties of the formose and the glyconitrile CRNs in chemical reactors such that information can be processed by the chemical systems. This concept requires the chemical input and the reaction conditions to be well controlled, which is done by automating all chemical actions in a scripting environment. Finally, ways to solve nonlinear classification tasks with the physical reservoir computer computing tasks with specific chemical action sequences have been developed, which are highly generic and allow for first computing tasks with chemical systems. The protocols developed allow analytical algorithms to be used for generation of feature and composition vectors in a machine-readable format. First examples for classification, complex dynamics and nonlinear predictions were presented and demonstrated to outperform tailored machine learning (ML) models, in particular for nonlinear classification. This ML tool implementation marks a significant step toward data-driven exploration and prediction within CRNs, providing a scalable solution for processing complex experimental data. The establishment of a pipeline for ML-enhanced reservoir computing, more complex computing tasks can be envisioned in the future.
CORENET will have a significant impact in various areas of systems chemistry, CRNs and reservoir computing (RC), and it has been progressing in line with the planned outcomes.
Looking at the formose CRN, precise data, analysis and important new concepts in reaction pathway self-organization are of high relevance to the prebiotic chemistry community. The framework developed using the formose reaction as a model can be transferred to other CRNs, allowing the rationalisation of complex reaction outcomes and the inspiration of detailed hypotheses for origin of life scenarios that consider the dynamic and out-of-equilibrium properties of the prebiotic environment. With an established reservoir computing capacity of the formose CRN network, we can now demonstrate a range of complex non-linear classifications. The fabrication of integrated Si-based CSTRs has scientific impact as it allows for further integration and robust and reproducible data collection. The fabrication methods include possible patentable inventions with respect to embedding electrodes and heating elements.
Looking at the formose CRN, precise data, analysis and important new concepts in reaction pathway self-organization are of high relevance to the prebiotic chemistry community. The framework developed using the formose reaction as a model can be transferred to other CRNs, allowing the rationalisation of complex reaction outcomes and the inspiration of detailed hypotheses for origin of life scenarios that consider the dynamic and out-of-equilibrium properties of the prebiotic environment. With an established reservoir computing capacity of the formose CRN network, we can now demonstrate a range of complex non-linear classifications. The fabrication of integrated Si-based CSTRs has scientific impact as it allows for further integration and robust and reproducible data collection. The fabrication methods include possible patentable inventions with respect to embedding electrodes and heating elements.