AI is replacing costly trial-and-error in developing novel biomaterials
DAN*NA, a Barcelona-based green bio-engineering company, has been attracting international recognition for its high-tech biomaterials. Its flagship product is a bio-PLA that is more flexible than market competitors while retaining transparency. It has also patented a new bio-based materials for tissue regeneration and bioprinting.
DAN*NA’s reputation is not just the product of human ingenuity. Its AI materials design software informs every stage of its development process. By mocking up a virtual ‘clone’ of target materials, developers can predict the physical properties of novel formulations without arduous experimental runs. DAN*NA’s system and others like it help developers identify how bio-based materials should be manipulated to match petrochemical materials on performance and properties.
The company’s computational approach to bio-materials development has driven it to the forefront of biomaterials innovation. Over the last three years, it has acted as a leading partner in several major private and public projects, both in Spain and aboard. One is the $6.9 million European Union Horizon 2020 Research and innovation programme Catco2nvers, which is creating new value-added chemical products using captured carbon emissions from the bio-industry. DAN*NA is also involved in BIOCON-CO2, another project developing methods of carbon conversion for biomaterials manufacture.
Machine-learning in materials design is not new, but its commercial adoption is widening. Worldwide patent applications for AI-driven material science rocketed from around 1100 in 2011 to over 3000 in 2017. Although AI tech for innovation and discovery is already widely used in pharmaceuticals, the technique is migrating into industrial biomaterial design.
What is the place of AI in biomaterials research?
The role that AI could play in the sustainability transition has been much discussed. Some believe the technology could guide environmental policymaking. Others say it could be deployed as a calculative aid for resource optimisation or to identify more efficient industrial processes. One of the most substantive ways that AI software can support economy-wide decarbonisation is by generating new, sustainable, and economically viable innovations in material design.
Many product applications have strict requirements on the physical properties that its materials must deliver. In food packaging, plastic containers can keep moisture out and heat in. Construction materials must not corrode, crack, or buckle under weight. Despite huge technical advances in biomaterials design made over the past two decades, some renewable materials still fall short of being perfect functional substitutes for legacy petrochemical derivatives. Because biomaterials are usually more expensive to make than their petrochemical counterparts, matching (and ideally going beyond) the performance of legacy materials is essential for capturing a wider market.
The physical properties of any material rest on their unique chemical compositions and molecular structures. Sometimes, how the molecular or nano-scale structure can matter more in determining mechanical behaviour than the elements that make them up.
However, there are a dizzying number of possible molecular combinations. This is motivating researchers to adopt AI to generate promising configurations.
IBM, AI, and sustainable computer chips
IBM are pioneering AI design for sustainable components in microchips. ‘Photoacid generators’ (PAGs) are a photosensitive material used in manufacturing computer chips. The chemical formulas of current PAGs are environmentally toxic so that computer industry is looking for greener alternatives.
PAGs are what allow tiny chips to etched with maximum precision. Ultraviolet light is shone through a mask cut in a wiring pattern. When the light pattern hits a microchip coated in PAGs, the PAG exposed to light will decompose into acids that eat the chip’s base material away, leaving a perfect pattern behind.
The development process for greener PAGs is costly, particularly if it is guided only by trial, error, and human intuition. Instead, the researchers are using AI to search for and combine sustainable materials into a novel kind of PAG that is both made from renewable materials and has all the right optical properties. The team used IBM’s Deep Search AI to trawl scientific papers for candidate organic materials. Next, researchers input these into IBM’s generative ‘Intelligent Simulation’ AI. Using these, the software offers suggestions for how these materials could be structured at a molecular level to achieve properties that make for effective PAGs. Another IBM technology then explore which of these outputs could function best.
BioBTX and AI for bio-chemical scaling
IBM also uses machine-learning to pick the best methods for actually making the PAGs. Another company that has used AI to find optimal methods for biomaterial production is German startup BioBTX.
BioTX wanted to find cost-effective production methods for bio-based aromatics ingredients. Aromatics are a group of chemicals used in clothing, pharmaceutical, cosmetics, computers, wind turbines, paints, vehicle components, and sports equipment. Currently, aromatics are made from petrochemical derivatives but BioBTX sought circular and biobased renewable versions. It found a way to break down a waste product from biodiesel production, glycerine, into three key chemicals inputs for aromatics: benzene, toluene, and xylene.
The bio-based versions of benzene, toluene, and xylene must be functionally indistinguishable from their petrochemical versions. However, any method for making them must also be cost-effective. The problem was that there are around 5 million potential methods for converting the glycerine into the three aromatics ingredients. Each use different raw materials, catalyst types, and temperatures. It is not obvious which would be economically viable and physically optimal without trying each one. Checking even a fraction of the possibilities in the lab would be prohibitively expensive.
To sidestep the problem, BioBTX teamed up with the University of Groningen on an algorithm that simulates real experiments on glycerine degradation methods and predicts their outcome. Based on the suggestions, the company set up a pilot plant to produce its first runs of bio-based aromatics compounds. Now, are working on constructing a full-scale factory scheduled to come into operation by 2023.
The chemical combination problem also wracks companies that work at the intersection of green chemicals and synthetic biology. Arzeda, a University of Washington synbio spinoff established in 2008, created a protein design platform to predict the properties of fermented chemicals. What’s unique about their software is that it has been able to generate entirely new molecules with properties not found in any existing synthetic or natural materials. The company takes these custom-designed molecules and produces them at scale through bioreactor fermentation.
California-based Zymergen, which went public in April 2021, are also melding machine learning with bio-engineering and bio-manufacturing. Using machine learning, they can learn how high-value chemicals cultivated inside the bodies of microbes will perform in an actual application. This is useful for selecting which strains should be taken into the scaling stage. Zymergen used this process to manufacture their bio-based electronic films.
Zymergen also use their software to identify how the genetic strains in microbes associated with better bio-manufacturing outcomes, like higher yields. Computer simulations can also give an accurate picture of trade-offs might be associated with optimising a given microbial trait. For example, a small alteration in DNA might produce a microbe that needs less sugar inputs but might take longer to produce the target chemical.
AI-based bio-design is at the heart of Zymergen’s business model. The company sells their simulation services to clients on top of using them to develop and manufacture new products in-house.
Industry and academia
Computational chemistry and industrial applications of machine learning are emerging fields. The biomaterials sector is also still in its infancy. As a consequence, industry-academic collaborations are being forged to refine software and expand the material databases to train them on.
In 2017, Toyota Research Institute (TRI) and Northwestern university, embarked on the Accelerated Materials Design and Discovery (AMDD) project to apply AI to advanced materials research. The programme sought out new materials for a lower-carbon automotive industry. In the first four years, it managed to predict 19 entirely new materials using a software developed by Soicheia Inc. The project drew on Soicheia’s gargantuan database of more than 200 million nanomaterials. The researchers generated nanoparticles with different compositions, structures, sizes, and shapes by feeding the data into a machine-learning algorithm equipped with basic physical and chemical principles. To date, it has published over 150 academic papers on lower-carbon battery and fuel cell materials.
The Toyota Research Institute project received an initial investment in that year of $35 million for four years. The project received a further cash injection of $36 million in 2021 to continue their research with several universities around the world, including the California Institute of `technology, Carnegie Mellon, and MIT. Professor Yang Shao-Horn who is participating in the research commented, “We’re working with TRI to bring together polymer synthesis, rapid robotic experiments, molecular simulation, and AI to establish new design rules for polymers”.
Machine learning offers a route to increased efficiency in basic research for innovative materials. This is critical for the biomaterials industry looking to overcome market perceptions about sub-par bio-based products. However, the uses of machine learning go further than quickly predicting the physical performance and economic viability of molecular combinations. It could give the bio-based sector an edge over the synthetic chemicals industry in generating entirely novel materials with unique properties that are difficult to derive from petrochemicals.