Periodic Reporting for period 3 - SMART-POM (Artificial-Intelligence Driven Discovery and Synthesis of Polyoxometalate Clusters)
Reporting period: 2018-11-01 to 2020-04-30
SMART-POM is important as innovation in chemistry across the world aims to drive down costs and to create efficiency in discovery and sustainable manufacture, which are considered cornerstones for sustainable industry grown and profits. This often translates to offshoring and centralisation of supply chains, but there is a vast untapped potential for the EU to leverage existing clean technologies to reduce cost and increase value through digitization, and crucially, cutting the cost of discovery and digitization. Also, the manufacture of chemical products, whether they be bulk, fine or speciality chemicals, such as APIs, is currently based on a model whereby a central plant is exclusively designed for the manufacture of the product, or a small range of products. This results in safety issues around both the storage and transport of such materials as well as the issues inherent in the large-scale manufacture of chemicals. Transport also introduces both cost and environmental footprints, and large FMCG companies, for example, are already exploring methods for decentralizing manufacturing. Large plants are often at the mercy of complicated and global supply chains of raw materials, the failure of which at any point will reduce or halt the capacity of the plant to produce materials and deliver them effectively. We will develop new techniques and methodologies as part of our drive to produce an easy to use system for the translation of traditional laboratory batch synthesis into an automated cartridge / modular system.
The key aim of the programme, to establish the new discipline ‘digital-chemistry’, via the use of digital systems in chemistry has not changed and is on course according to our work plan. To develop a new synthetic chemistry and engineering platform for the discovery of molecules, clusters and nanomaterials requires the development of several integrated hybrid system integrating wetware (chemical reagents), hardware (reactors and sensors) and software (intelligent algorithms). In initial studies the proof of principle has been completed. That is, by ‘digital’ programming we have been able to optimise / change the course of the wetware as a function of the properties measured using algorithms controlled with a software system. The research is multidisciplinary requiring the expertise of a team of chemists, electrical engineers and physicists, who share the vision of integration and advanced software control of matter.
i) Construct a universal building block library of POM sub-units, clusters and methods for their programmed synthesis & assembly into super-structures with designed functions and applications including dynamic approaches to the assembly and synthesis of new systems.
Here we explored the derivatisation of metal oxides. This is important for many reasons, for example, because transition-metal oxides are reported to improve power conversion efficiencies in organic photovoltaics and have been used as low-resistance contacts in organic light-emitting diodes and organic thin-film transistors. To explore the development of new materials, it was important that we selected POM cluster material precursors that could be easily synthesized and derivatised, which is vital if POM-clusters are to be linked together in a given topology to yield framework materials. Derivatisation is difficult since the synthesis of polyoxometalates is commonly done under one-pot conditions, and often even simple synthetic conditions lead to a complex mixture to be formed which is mainly governed by self-assembly mechanisms. This implies that a slight variation in the reaction conditions (change of pH, temperature, etc.) can easily give access to several different cluster architectures. In some cases, chance seems to govern the self-assembly process and despite general speciation rules having been observed, the resulting architectures cannot be predicted often. Those structures can, however be utilized once known as building blocks in other materials. In fact, the reliable synthesis of complex POM clusters is an important enabler in the discovery of new materials.
We were able to design catalytically active systems or multi-functional materials that can have several roles ‘encoded’ into a single compound. This was possible because a cluster capable of redox activity, coupled to a cluster capable of small molecule binding, and with a cluster capable of photochemical activation was shown by us be an ‘allin-one’ functional material. The design was done by selecting wheel-shaped POMs, metal ion linkers, and we engineered the assembly in an efficient and high yielding manner. ALso, the ability to build these materials was critically limited by the obscure nature of the synthesis of many POMs. However, thanks to recent advancements during our grant regarding mechanistic control, and system automation, it is now much easier to reliably build such building blocks, whereas before this is a big barrier. This enhanced synthetic access, combined with the new building block libraries, and emerging taxonomy, has established polyoxometalate frameworks as a new class of metal oxides.
ii) Establish the concept of using algorithms, feedback control, and in-line spectroscopies to follow self-assembly and explore reaction parameter-spaces for novelty. This will use reactor systems including multiple connected ‘one-pot’, linear, networked, and arrayed flow-reactor systems for the programmable assembly and scale up of the cluster synthesis.
Understanding the supramolecular self-assembly of complex inorganic molecules poses a difficult problem since it relies on two contingent events. To make a discovery, one should first find the conditions under which the building blocks assemble and then the conditions under which the product aggregates into crystals to be isolated and characterized. The vast number of combinations of the experimental conditions and the coordination modes of the transition metals taking part in the building blocks means that a full exploration of the chemical space of any given compound would be impossible. For these reasons, the intuition of experienced chemists is required to design the appropriate experiments in order to determine the right conditions for the isolation of any new products. But intuitions can be biased by both the current knowledge on the field and the frame of mind of the experimenter - making important discoveries difficult to achieve. In this work we designed and investigated a new approach for probing the envelope of both the synthesis and the crystallization process of a new range of polyoxometalate compounds incorporating Lanthanides. Our method is drawn from recent advances for active data acquisition in the field of machine learning, known as active learning. Active learning consists of methodologies able to decide what experiments to perform next in order to optimally improve the understanding of the system at hand. We compare our algorithmic method with a random screening process in the exploration and modelling of the crystallization conditions of compound. Importantly, we study how human experimenters approached this specific problem and compare their strategies and performance to our machine-learning approach.
iii) Use the universal building block library within the reactor system to discover new clusters and cluster based materials to elucidate the underlying mechanism of assembly. These experiments will also explore the application of the new nano- molecules and materials as catalysts, and devices. This will then set the stage for applied work with industry or spin-off companies.
We have used a flow reactor system approach to both explore the mechanism, and the synthesis of complex polyoxometalate clusters. For example, by using the flow system we were able to generate a stationary kinetic state of the ‘intermediate’ molybdenum-blue (MB) wheel, filled with a guest to give a host-guest complex. The MB host−guest complex assembly was carried out under controlled continuous flow conditions enabled selection for the generation a major product, and allowed the reproducible isolation of this host-guest complex in good yield, as opposed to the traditional “one-pot” batch synthesis which typically leads to crystallization of the. This is of crucial importance since further reduction of the wheel results in the expulsion of the guest indicating why this was not observed before reproducibly. Also, further experiments showed an increase in the yield and the formation rate of the wheels by deliberate addition of preformed to the reaction mixture. Dynamic light scattering (DLS) was also used to corroborate the mechanism of formation of the MB wheels through observation of the individual cluster species in solution.
During the next few months we will develop new machine learning systems to allow the seeding of clusters with templates. Since the assembly of nanomaterials from the top-down gives precise structures, it is costly whereas bottom-up assembly methods have to be found by trial and error. Nature evolves materials discovery by refining and transmitting the blueprints using DNA mutations autonomously. Genetically-inspired optimisation has been used in a range of applications, from catalysis to light emitting materials, but these are not autonomous, and do not use physical mutations. In the next set of work we aim to develop an autonomously driven materials-evolution robotic platform that will allows us to reliably discover the conditions to produce molecular nanosystems that can run for many cycles, discovering entirely new systems using the opto-electronic properties as a driver. Not only do we aim to reliably discover the digital-material-code to synthesise these materials, we will try to seed in materials from preceding generations to engineer more sophisticated architectures. Over at least three cycles of evolution we will explore how the seeds from each generation can produce new systems.