Synthetic biology (synbio) has been a recent darling of venture capital funding, raking in billions and generating multiple recent IPOs. Moreover, synbio startups have demonstrated modest success, and some of them are on a fast track to becoming the next generation of sustainable chemicals companies. Synbio, which combines the use of artificial intelligence (AI), gene editing, molecular biology, and fermentation, has high potential to become an alternative way of manufacturing chemicals, with numerous advantages, including better reaction rates, selectivity, and higher yields. Synbio is also relatively similar – at least on the surface – to materials informatics (MI), which uses AI to help design and invent new materials, formulations, and chemicals. Despite the surface similarities, there's been much slower progress in MI, with limited commercial adoption for dedicated startups and certainly no multibillion-dollar boom in funding. We think the issue is the difference in the business model; nearly all MI companies are on a service provider, software-as-a-service (SaaS) model, and successful synbio companies are building their own products.
The service provider model:
Most MI companies started with project-based services with no user interface (first generation). As they grew, many of them (Uncountable, Aionics, AI Materia, Intellegens) developed a software platform for customers to perform their own data analytics. The MI companies charged a fee for platform use (based on tokens, projects, offices, or monthly/annual fees – second generation). While they hoped to achieve SaaS-type high growth, the unfortunate reality is that customers did not scale up for two main reasons. First, most customers were R&D units from chemicals companies, and all of them had data security concerns (while this is mostly justifiable, one can argue that they have sometimes been overly conservative). This limited the chemicals companies' willingness to put data on these platforms and thus limited their utility. Moreover, the platforms only targeted R&D, thus limiting their potential customer base to a small slice of the chemicals industry. Perhaps learning from these experiences, Citrine has emerged to be a data infrastructure provider (third generation) on top of the other services (though it does not provide project-based services anymore). The goal is to help R&D customers restructure their own data infrastructure to make it more suitable for data analytics. We argue the data-infrastructure-as-a-service model is the best a startup can do under this service provider track, and the next step is to augment the platform with more downstream features. That said, there is limited room for this type of infrastructure company, as a couple of them (or even one) can, in theory, serve all chemicals companies.
The build-your-own-product model:
Zymergen is arguably one of the most successful synbio companies, and it is on this model. It builds a metagenomics database for microbial strain engineering; once it engineers a strain that can take the desired feedstock and generate the targeted product via fermentation, it applies high-throughput screening to learn the right conditions for scale-up fermentation. It has so far launched one product and has 10 more lined up. Genomatica, another successful synbio company that just announced a fundraise, takes a different approach but can also be considered a product company. Rather than developing new materials, which has the associated risk in terms of market acceptance but also the attractive high-profit margin, Genomatica replaces the existing petrochemical-based processes with a bio-based approach. While it specializes in fermentation scale-up and is still primarily on a license-based business model, it has also launched its own brands. While we use these two companies as examples, there are many more where those companies target other markets, such as pharmaceuticals (Antheia and one can argue even Recursion Pharmaceuticals), food (Impossible Foods), and other consumer products (Spiber). While the market for service providers is limited, product companies' market is much bigger, and one can argue that the limiting factor is the boundary of the technology.
What then, should MI practitioners do?
They should develop their own products, though it will be hard for existing "service provider” companies to switch to building products. As they plan to develop materials, they will need to shift their IP strategy to owning the key patents and thus will risk losing some platform customers. That said, here is a general guideline.
Materials data in the public domain is useful but limited, so companies on this route need a way to generate data (access to high-throughput screening or computing power for simulated data). Another factor is the patent landscape – owning IP is key, and companies should identify patent white space and start from there. Moreover, given the notoriously lengthy and difficult process of growing the market for a new material, companies need to "engineer a new market" from the beginning. It is critical to show that this approach works on some low-hanging fruit. Once companies can develop a new material without any data other than their own, the value proposition is valid, and they can patent the material and license the technology to manufacturers. There are a few MI companies taking this approach (Phaseshift, QuesTek Innovations, and Alloyed, formerly OxMet Technologies). Among these, QuesTek has made some good progress on alloys for niche applications. The relative ease of manufacturing advanced alloys, existing small-scale production for high-value applications, and good data on metals all contributed to this success; a polymer-focused startup with the same business model, such as Materialize.X, faces much more substantial barriers. Optionally, companies can go further and build downstream products using the developed materials and even manufacture those products directly (existing manufacturers may not have the capability required). Metamaterial has demonstrated great success in this route.
What should chemicals companies do?
Companies in the specialty chemicals and materials industry are much better-positioned than MI companies. They have data and resources and are much more financially stable. Therefore, they should actively incorporate MI into the product development process. A key first step is organizing R&D data into new infrastructure. Trade-offs should be made on a case-by-case basis, such as converting existing data with the old format to a new format vs. collecting new data from scratch, collecting data through experiments, or simulation. Integrating R&D with product design, manufacturing, and sales units is the longer-term vision. The idea is to optimize a material recipe based on the required features of all these units. Evonik and Dow have some early approaches to connecting R&D with sales, but to the best of our knowledge, no company has integrated these units on a systematic scale yet. To conclude, while startups are not realizing the full potential of MI yet, the success in synbio does provide us a version of MI's future. Chemicals companies should actively integrate MI into their product development process. At the same time, they should be aware that some MI service providers (existing or new) could become their competitors in five to 10 years.