WE PONDER THE QUESTIONS: "How do molecules compute?", "HOW DO MOLECULES LEARN?", AND "How do molecules transition into life?"

Keywords: Systems Chemistry, Out-of-equilibrium Networks, Complex Systems, Nonlinear Dynamics, Molecular intelligience, neuromorphic computing.


How can CRNs enable novel concepts in artificial intelligence?

Rationalizing the ability to learn, and the capacity to adapt to an environment, goes beyond current synthetic designs. Artificial neural networks can perform intelligent tasks (and in some cases outperform the human brain). Intelligence of this sort, however, require enormous amounts of energy and new types of hardware, with different energy requirements as compared to conventional silicon-based components, are urgently needed. Our current research focuses on the development of platforms (based on microfludics or hydrogels) capable of creating networks with tunable and scalable chemical interactions. Our projects will start from established enzyamtic- or thiol- reaction networks and explore the possibility of translating simple bistable systems into complex multistable systems. They exploit the dynamic exchanges of networks of chemical reactions with their environment to create spatial and temporal interactions with high redundancies (thereby capturing vital features of artificial neural networks in CRNs). The simple means employed to create and build up so-called hidden layers will make the design of CRNs amenable for novel materials with the ability to learn and adapt.


History Dependence in an Enzymatic Reaction Network. 

We provide novel methods to control systems’ dynamics, by changing how stable and transient behaviors of CRNs can respond to gradients. We recently demonstrated that a simple CRN is capable of dynamic switching between different states. Briefly, our network comprises a positive feedback loop (in which trypsinogen, Tg, is converted into trypsin, Tr). and an inhibitor (I) which can suppress the activation of the autocatalytic reaction. Under continuous flow conditions using a so-called continuously stirred flow reactor (CFSTR), Tg acts as a fuel for generating the catalyst Tr and its process is regulated by the inhibitor concentration ([I]o). We developed a method ‘linear inhibition sweep’, (i.e., a molecular analogue for linear voltammetry  measurements) to probe the history-dependence of this network. Essentially, hysteric behavior can be found in steady states but upon increasing the slope of the gradient, the same bistable network was capable of various complex behavior such as synchronization, resonance and adaptation. This work, thus, reveals that a simple three-component CRNs is already capable of memristive behavior (i.e., function that emulates the working principles of neurons). 

We are examining if perturbation experiments under repetitive linear inhibition sweep conditions could reveal how CRNs of this type can be switched on and off. These studies can be easily and logically extended to configurations comprising multiple bistable systems in series-, parallel- or array-coupled CFSTRs to examine how information is transferred between CRNs.

Selected publication(s): Kriukov, D. V.; Koyuncu, A. H.; Wong, A. S. Y.* "History dependence in a chemical reaction network enables dynamic switching" Small 2022, 2107523. 

Local Chemical Feedbacks enable Artificial Intelligence. 

We are establishing a microfluidic method to create spatially-organized activation and inhibition feedback loops. Briefly, a thioester (in yellow) and a cystamine (in cyan) together can create an autocatalytic production of thiols. Crucially, by incorporating localized reactions, this work will create the possibility to inhibit the autocatalytic events or activate new interconnections at various times and locations on a microfluidic chip (dynamically changing of the number of types of thiols, thus, create the hidden layers).

Capturing Higher-order Interactions in Chemical Reaction Networks.

Recent advances in network science suggest that pairwise interactions, a distinct property of graphs, may imply an inaccurate representation of real-world network interactions. Chemical Reaction Networks (CRNs) are different: They comprise non-pairwise interactions. Could this inherent nature of chemical reactions enable modelling of complex higher-order interactions? In this project, we combine the design of chemical reaction networks with probability analysis of random graphs to create a joint experimental and theoretical perspective on network complexity.  We are looking for two PhD students: one PhD with a background in chemistry, and another PhD with a background in mathematics. Please see our vacancy page for more information. 


What are design principles for chemical reaction networks? 

During my Ph. D. research at Radboud University, under the direction of Prof. Wilhelm Huck, Albert Wong developed strategies for designing chemical reaction networks and examining the robustness and resilience of reaction networks. The methods include the development of i) a protocol for the synthesis of small inhibitory molecules, ii) mathematical models based on differential equations for guidance, predictions and validation of experimental results, and iii) microfluidic platforms to maintain reactions out-of-equilibria (incorporate fluxes in the reactants). He demonstrated that networks of reactions can produce sustained oscillating concentrations (a well-established hallmark of out-of-equilibrium complex systems). This work enabled a translation of design principles of biological systems (that use regulatory motifs, and feedback loops) into a practical ‘chemical programming language’.

Selected publication(s): (1) Semenov, S. N. & Wong, A. S. Y. et al. “Rational Design of Functional and Tunable Oscillating Enzymatic Networks” Nat. Chem. 2015.(2) Wong, A. S. Y. & Huck, W. T. S. “Grip on Complexity in Chemical Reaction Networks” Beilstein J. Org. Chem.2017

How did molecules assemble into life? 

As Rubicon postdoctoral fellow Albert Wong worked with Prof. George Whitesides at Harvard University, where he branched off into examining how simpler but prebiotically relevant types of chemical reactions can self-assemble into complex, robust networks. He experimentally demonstrated that an environmental noisy environment may have stabilized, rather than destabilized, the emergence of dissipative networks of reactions. Albert and colleagues thus found that the transition from ‘molecules’ (or disordered chemical reactions) into organized ‘networks of reactions’ benefitted from, and in some cases required, non-conventional conditions (that is, out-of-equilibrium conditions, nonlinearity, heterogeneity, and noise) rather than conventional conditions (the quiet and uniform conditions of the laboratory) for optimum performance. This work sheds a light on the origin of chemical pathways that now form the basis of metabolism.

Selected publication(s): Cafferty, B. J.; Wong, A. S. Y.; Semenov, S. N.; Belding, L.; Gmuer, S.; Huck, W. T. S.; and Whitesides. G. M. “Robustness, Entrainment, and Hybridization in Dissipative Molecular Networks, and the Origin of Life” J. Am. Chem. Soc. 2019.