I’m interested in developing a technique to explore the emergent properties of biomaterials for technological purposes. Biomolecules are highly complex and possess a great deal of information. My approach is to use a machine learning system that explores the highly complex phase space of such high information biomaterials, such as lipids and protein mixtures, to find emergent behaviours that have a value. To achieve this aim I am in the process of developing a novel micro-fluidic additive manufacturing technique that will be controlled by a machine learning system. The result will be the ability to digitally control both the molecular details (via DNA) and the additive manufacturing process for exploiting emergent phenomena.
Imagine a world with finite material resources and an unlimited supply of energy, in which every atom is already being used for an economic purpose. The only way to increase economic value in such a scenario is to find efficient ways of re-arranging matter as demands change. Indeed, biology is essentially a system for rearranging matter efficiently. Instead of finding new kinds of raw materials to solve each new problem, biology has learned how to encode information into material (e.g. DNA sequences), that allow the material to be-retasked for new applications. Biology is a system for finding and remembering ever more useful ways of organising fundamental building blocks (amino acids) in response to changing conditions.
Furthermore, biology often arranges materials to solve multiple problems simultaneously by exploiting emergent phenomena. For example, the shape of a wing generates lift for free when it is placed in an airflow. One could easily take the materials found in a wing and re-arrange them in a way that cannot generate lift. Lift is a useful emergent property that arises from the shape of a wing.
Often one finds such interesting emergent properties in materials by combining multiple sub-systems in ways that prevent each component within the system from behaving as it does alone. Such an arrangement is known as frustration. A great example is the bauhinia seedpod in which an internal stress builds up between two bilayered sheets as the seedpod grows. Upon maturing that stress is released all at once, thus flinging the internal seeds over a wide area. Similar internal frustrations can control the morphology of amyloid fibres, and are partly responsible for the complex behaviour of collagen.
Such emergent phenomena are hard to predict and are often found by co-optimising material properties at multiple length scales simultaneously. Consequently, natural materials, such as bone, bamboo or wood, are often hierarchical and contain both long (1 m) and short range (1 nm) order, as well as order at many intervening scales.
My research then aims to use machine learning to control a microfluidic system containing recombinant protein and lipids to find useful emergent phenomena, by creating frustration in hierarchical materials. Such an approach will enable digital control over virtually any aspect of the material. My objective is to determine the limitations of this technique and to use it to efficiently explore phase spaces of complex biomolecules to find useful combinations of materials that perhaps even biology hasn’t had time to consider yet.
By applying machine learning to a micro-fluidic system we can create a dynamic feedback loop in which the full physico-chemical properties of a complex system are completely accessible to the top level system – whether we understand them or not. We can select target properties for our material, and allow our machine to evolve towards those properties.
The ultimate objective is to reduce the total number of materials in the global economy. Every new material we introduce for each new application has associated with it an entire pyramid of industries and a unique supply chain. Once we begin to mix materials in complex systems like smart phones or computers, we make the recycling task extremely difficult. My research is asking the question to what extent can we re-task a core set of materials merely by changing the information content of the hierarchical assembly – just like biology swaps amino acids around to create new proteins, and combines those into more complex assemblies.
Furthermore by limiting myself to the palette of materials used in biology, any materials produced this way will be inherently biodegradeable and fully compatible with existing biological ecosystems.
My hope is to demonstrate digital control over every level in the manufacturing process from molecules to complete systems. The result will be a powerful generalised fabrication technique, that is capable of using the same core materials (C, H, N, O, P, S and trace elements) as biology, and assigning economic value to complex combinations of those materials.
Such a sophisticated materials and manufacturing platform will form the basis of the next generation of self-assembling additive manufacturing and drive the conversion to a circular economy.