The chemicals industry, a cornerstone of modern life, faces a critical challenge: balancing industrial growth with environmental responsibility. By harnessing AI and biocatalysis, we can pave the way for a more sustainable future. In recent years, expanding biological data and advances in computational techniques have given rise to AI algorithms capable of everything from designing new proteins from scratch to predicting diseases from any possible mutation. These tools are already being broadly applied across drug development with their potential impact soon to emerge. Yet there remains an underserved application: their use across industrial biotechnology as a means to help clean up sectors that have polluted our communities and planet. The time has come to build circular economies and a better world for our children.  

The Hidden Toll of chemical Production

For more than 75 years production of chemicals has increased to now exceed 1T pounds annually in the US alone.  The vast majority of these chemical products are derived from oil & gas feedstocks using catalytic processes that are usually energy intensive and often generate toxic byproducts. Together, this production represents part of an ~$6T industry. But that industry faces increasing pressure from both consumers and regulatory agencies to reduce waste and increase circularity/sustainability. Together, leading chemicals producers are investing ~$7B each year in support of these mandates.

Across industries there exist diverse process waste streams from CO₂ or other gasses to biomass or plastics. Often, these are either simply vented into the atmosphere, buried in a landfill, or worse yet, finding their way into the natural environment where they can wreak havoc on plant, animal, and human populations. In the face of population growth, consumer demand, and sustainability goals, the pressure to decarbonize the chemicals industry has never been higher.

Revolutionizing Chemical Manufacturing with Biocatalysis

Over the course of at least the last decade, thanks in part to investments across the pharmaceutical industry, the cost of protein manufacturing has been significantly reduced. This has lowered both small scale costs for early design-build-test-learn cycles and for scaled commercial operations. At the same time computational tools have made striking advances in predictive power across biological applications thanks in part to growing genomic and proteomic datasets. Together, these have led to expanded abstraction and services offerings across the biotechnology sector which facilitates greater outsourcing and capital efficiency.

On a similar timetable, we’ve seen the advance of computational approaches in biology grow from flux balance analysis, molecular dynamics, and physics-based models to neural networks, large language models, and de novo design. Even more recently, researchers including Levin et alhave begun to apply classically chemical approaches like retrosynthesis to biology using AI (neural networks). Such efforts allow us to escape the confines of evolved metabolisms and the stitching together of heterologous pathways to achieve industrial manufacturing goals. Over the eons, natural biology evolved for growth within ecological niches, not optimal production of chemical products of interest. Thus, we set the stage for the potential to use enzymatic retrosynthesis search to map novel pathways/processes for chemical manufacturing. This unconstrained biotransformation space, loosely defined as all possible enzymatic catalysis multiplied by all known chemical intermediates, is not only non-navigable by the human mind but wildly undersampled to date.

Disrupting Markets: A Path to Circularity

First generation synthetic biology companies were often focused on biofuels production thanks in part to the giant market size carrot at the end of the stick, but as they moved towards commercialization the practicalities of commodity unit economics in the backdrop of contemporary biology tools became untenable. As the pendulum swung for the second generation, most focused on higher value ingredient products such as flavors and fragrances or cosmetics. While certainly easing unit economic constraints, ingredients faced a rather high spec threshold which challenged drop-in adoption and by extension revenue growth. Based on the aforementioned advances in technology as well as others, there exists an opportunity to pursue the goldilocks marketplace of ‘platform chemicals’. These are chemical products which tend to be manufactured in excess of 1M tonnes per year, sold in the $2-10/kg range, and utilized across industry verticals. Exemplary products might have uses ranging from pharmaceuticals, food, and cosmetics to de-icing agents, plastics, surfactants and lubricants. Their global markets are typically estimated in the >$1-5B range annually with domestic production typically exceeding millions of tonnes. The existing marketplaces for such chemicals are dominated by large multinational chemical companies using oil and gas component feedstocks and manufacturing processes that haven’t seen innovation in decades.

AI and Techno-Economics: Bridging Vision with Reality

Interestingly, the problem of waste generation/remediation and sustainable chemical production is often experienced by the same companies. Major chemical companies suffer from multiple and diverse waste streams that they are eager, nay obliged, to see mitigated and are simultaneously interested in the ability to develop new more sustainable processes for manufacturing. De novo designed biocatalytic pathways and cell-free reactor systems represent an opportunity for them to potentially do both at once. Elsewhere, product end users lament current supply chain dynamics and desire (sometimes as an extension of their own customers’ desires) more sustainable sourcing. For oil-derived platform chemicals, indeed there are many potential end users who may not have the internal or customer imperative for sustainability but whose adoption thereof will benefit society.

Assuming a $4/kg unit price for target product, then a first $1M in revenue could be realized upon production of 250 tonnes. This scale might be achieved initially in a pilot facility with reactors on the scale of 5-10k liters over a longer time frame (and at lower margins) or in a higher capex facility similar to that modeled by Singh et al wherein their assumptions allow for achieving such volumes on a weekly basis. In either scale setting, AI algorithms are already being developed to integrate diverse process parameter data streams to help optimize product yield/productivity and thus margins. Prior to fundraising and capital investment in production facilities, target products should be thoroughly vetted through primary market research with prospective customers to secure letters of intent or perhaps even off-take agreements if product samples are available through benchtop validation. Early adopters at even higher price points can be initially serviced by pilot facility production to begin generating revenue as production ramps up.

A Roadmap to Decarbonization

Irrespective of advances in computational biotechnology and synthetic biology, the future catalytic potential/efficiency of AI-predicted, non-natural enzyme/substrate pairings will be foundational to success or failure in these approaches. As Bar-Even et al described years ago, nature tends to evolve only moderately efficient enzymes. And while our practitioners have decades of experience focusing and improving that efficiency using classical enzyme engineering approaches, some of the latest AI tools like LigandMPNN have improved effectiveness.

For each specific feedstock-to-product biotransformation, detailed techno-economic modeling initially anchored in assumptions can help set benchtop bioengineering goals and process optimization on the path to commercialization. Waste feedstocks of all kinds at varying price points are foundational to that modeling but also geographically fragmented. Future success in this regime will depend not only on securing product demand but also on aligning that demand with regional feedstock supply and tightening supply chain logistics.

In the not too distant future, biology may indeed eat everything, but we need to help it make sure that, at least for this meal, its eyes aren’t bigger than its stomach.

– Guest Blog Post by Christopher M. Pirie, PhD