TechBio in Industrial Biotechnology
How Machine Learning Can Overhaul the Physical Goods Economy
Quick Summary
The ‘TechBio’ approach combining bespoke Wet Lab systems with ML-driven ‘Dry Lab’ techniques has been well-documented as a platform approach to product development in the therapeutics space, but is not as well-trodden in the industrial biotechnology world.
There is opportunity for this ‘TechBio’-style approach across each of the three key stages of the industrial biotechnology product development process, from Molecule Selection, to Strain Design, to Manufacturing.
The largest opportunity exists in the development of vertical-specific TechBio-style discovery engines to characterize and discover useful new molecules for applications across all physical good industries from food to cosmetics to materials and more. These discovery engines represent a tool to build scalable portfolios of multiple products within a given category or industry vertical.
Much like the rest of the world, the biotech community has embraced the promise and buzz of all things machine learning over recent years.
In the world of traditional biotech and therapeutics, this tech-meets-bio approach - TechBio - has become well-defined and the business models for success well-blueprinted (I’m going to mostly refrain from characterizing the TechBio approach in this post, but some good background reading is this post by Elliot Hershberg, this blog by Amee Kapadia, and this 2019 podcast by Nan Li and Zavain Dar which was an eye-opener for me personally starting off my career in the world of biotech around the same time). In one sentence, the TechBio movement involves iterative loops of wet lab empirical testing and dry lab Machine Learning (ML)-driven analysis to create virtuous cycles of ever-improving discovery for new therapeutic products such as small molecules, antibodies, or gene therapies. Massive thousand-or-million-large libraries of interventions -- libraries of small molecules or antibodies or genetic knock-outs -- are created and tested, the results of those tests are fed into computational algorithms to help identify the factors behind the most successful hits, and the algorithm then helps compose the next round of library construction and testing.
This general formula [Wet Lab] X [Computation] X [Recursive Iteration] = [Discovery] can now be copy-pasted across any number of new modalities and biological interrogation mechanisms. LabGenius has a bespoke platform for antibodies, Dyno Therapeutics for AAV capsids, Enveda Biosciences for natural products, Atomic AI for small-molecule targeting RNA . . . the list goes on. Recursion Pharmaceuticals, Eikon Therapeutics and others have each built their own platforms based not on a specific modality, but on unique assay and biological interrogation platforms.
While these TechBio approaches are relatively well-trodden in a therapeutic discovery context, they are far less well-traveled in the context of industrial biotechnology product discovery and company building. These same toolkits will have a massive impact in consumer and industrial biotech, tackling weighty areas across food, agriculture, chemicals, climate and more.
TechBio in Industrial Biotechnology
Industrial biotechnology refers to the use of the biotechnology tool-kit (including newer, synthetic biology approaches) to manufacture products across a wide range of physical goods industries (for a primer on industrial biotechnology and some of the industries that will be impacted, you can check out this McKinsey Report, the SynbioBeta website, or just peruse the Ginkgo Bioworks Customer Case Studies). Industrial biotechnology companies generally come in two shapes: (1) companies making existing products better, cheaper, and more sustainably or (2) companies creating completely novel products and ingredients. Industrial biotechnology has already - often behind the scenes and unknown to the end consumer - brought many new products and ingredients to all physical good industries, from food to agriculture, to chemicals, textiles, plastics, and more. This will only accelerate in the years to come, and this TechBio approach has a role to play in that acceleration.
No matter the target product, there are three big-picture steps in bringing an industrial biotech product to the world:
Molecule Selection: Working backwards from a customer need to a desired function to a precise target molecule for a given application.
Strain Design: Designing a microbe or enzyme to make that target molecule;
End-Product Manufacturing: Scaling the manufacturing.
While each of these three steps will benefit from adoption of ‘TechBio’ approaches (Remember: [Wet Lab] X [Computation] X [Recursive Iteration]), the selection of new functional molecules and ingredients is a key barrier to industrial biotechnology living up to its promise to re-invent our physical world. Biotechnology has expanded the physical building blocks available to society with a palette of complex new molecules that have never been available before via petrochemical or agricultural methods - from this new molecular tool-box, how can we find the needle-in-a-haystack ingredients that can drive innovation in food, human health, or materials?
The mapping of molecules to functions and unmet customer needs across verticals will be a ripe area for technology and company building in the years ahead. Strain engineering today is still a relatively expensive and time-intensive undertaking, so intelligently choosing the right products and molecules to target before launching expensive strain engineering is crucially important.
TechBio in Molecule Selection
The discovery of molecules or other biological outputs for use as products within a specific application has been perhaps the largest hurdle for the field of industrial biotechnology in recent years. In a world where we can grow anything, how the heck do we decide what to grow?
Within every application vertical, entrepreneurs and researchers have the opportunity to take the [Wet Lab] X [Computation] X [Recursive Iteration] approach to identify new molecules and new ingredients with unique and useful properties. Scientists and inventors across food, agriculture, chemicals, materials, and more have never had access to the palette of molecules and building blocks that biology makes accessible. With this wider ingredient quiver, there is a glaring need (and opportunity) for ML-driven approaches to help ID the most promising molecules for a given application.
Some of these discovery workflows will themselves utilize synthetic biology approaches, while others will utilize synthetic chemistry or other tools; however, all must be fluent in and collaborative with biotechnology in order to understand what molecules are actually economically compatible and feasible to produce and scale using biotechnology. For example, a company seeking to characterize the food functionality of proteins is likely to use synthetic biology workflows to produce protein libraries for its Wet Lab data collection, whereas a company focusing on novel chemical polymers may opt to use synthetic chemistry for more high-throughput sample production at the lab scale. At other times, synthetic biology techniques will allow for the development of novel functional screening technologies - we’ve previously written about the power of Octant’s GPCR screening platform with potential applications not only in therapeutics but also in flavor and fragrance. In cases like these, the synthetic biology tool-kit plays a key role in both the molecule selection workflows and, later, the development of strain production platforms for the selected molecule.
Ultimately, these discovery workflows will ID compelling molecules and ingredients. Once these are identified, they can be passed over the fence to strain engineers to enable cost-effective and scalable production. With the arrival of extremely powerful, general-purpose strain engineering platforms such as Ginkgo and others, this product discovery capability is where I’m most excited about the use of ML in industrial biotech. As the actual synthetic biology and strain engineering becomes democratized into open platforms like that of Ginkgo Bioworks, the value will increasingly shift to these companies with product discovery - ‘molecule selection’ - engines. Much as therapeutic TechBio companies are rinse-and-repeating this TechBio approach across new drug modalities, this brand of Industrial Biotech TechBio company will be modularly built to disrupt each narrow vertical of our physical goods economy.
Of course, just selecting the right molecule for a given problem is only the first step of successful product development and company-building. To succeed, companies will have to pair these ML-driven molecule selection platforms with strong execution in identifying the right problems to solve, in developing customer-ready products, and in building economical manufacturing operations.
Vertical-Native TechBio Discovery Engines
These discovery engines and companies will be built out across many unique application areas and verticals. Each will require a unique wet-lab / dry-lab interface to build the datasets required for their application space. We’re in the early days of this approach which requires fluency in both synthetic biology and the given application area (i.e., food formulation, material science, agricultural science), but companies across diverse end-industries are already pioneering this model with vertical-specific, ML-driven molecule discovery engines that cross-talk with industrial biotechnology production platforms. We’re seeing this technique used at companies like Motif Foodworks to identify proteins with a desired food functionality (in a specific pH and temperature environment) like gelation or mouthfeel, to identify polymers with unprecedented material properties, and to find sweet proteins that actually taste good.
Crucially, the screening technologies needed will very greatly across application - screening for food ingredient organoleptics and mouth-feel will require a different workflow than observing an agricultural pest repellant’s efficacy or a new polymer’s strength and transparency - this need for multiple bespoke platforms will be even more profound than in the therapeutics space where common infrastructure and methods (i.e., phenotypic images or cell-model cytotox screens) may be leveraged across therapeutic modalities.
TechBio in Strain Design & End-Product Manufacturing
There is also ample opportunity for TechBio style approaches to further improve the remaining two steps in the industrial biotechnology workflow - Strain Design and End-Product Manufacturing - and we’ll look briefly at those here, without getting too deep today.
Over the past decade or more, computational approaches have had massive impact on strain design and engineering. Very recently, there has been exciting progress in ML-guided workflows ranging from full pathway assembly (See Amyris’s 2023 paper on their Metabolic Engineer AI Lila), to enzyme and protein engineering, to ML-guided directed evolution workflows. In the generative AI world, companies like Profluent are even creating sequence-novel, fully de novo enzymes for various applications.
Despite the technological ingenuity, I worry companies developing solutions for just one step in the strain development workflow will struggle to capture sufficient value (at least in industrial biotechnology, therapeutics being a different animal). Value capture and unit economics for synthetic biology platform companies are notoriously difficult, and providing a single technological point-solution rather than full turn-key strain development may prove difficult footing on which to build a venture-scale business. That’s part of the reason we’ve seen larger companies scooping up great synbio point-solution technologies, such as Ginkgo’s acquisition of Altar & F-Gen and Inscripta’s merging with Infinome and Sestina. Companies bringing new platform technologies will be pressured to either merge into these larger platforms or attempt a perilous pivot to become product companies.
In scale-up and manufacturing, the creation of representative scale-down systems such as the Ambr250 (benchtop, 250mL disposable fermenters) have made medium-scale DoE and data-driven approaches beneficial and often highly predictive of processes at scale. Companies like Culture Biosciences and Pow.Bio have democratized access to ML-driven fermentation optimization that previously was only practiced in-house by stalwarts of industrial biotechnology such as Novozymes, Cargill, and DSM. That said, the challenge for companies scaling their products today is often more focused on DSP processes and recovery yields, and the role for massively parallel experimentation and TechBio-style approaches here remains far more opaque.
The Wrap-Up
There are many opportunities for the TechBio approach ([Wet Lab] x [Computation] x [Recursive Iteration]) to add value across the three steps of industrial biotechnology product development: Molecule Selection, Strain Design, & Manufacturing.
The largest value opportunity lies with companies building vertical-specific Molecule Selection engines - these companies will be the ones bringing totally novel products and ingredients into our world. Value will continue to shift to these application-focused companies as strain engineering becomes increasingly democratized via large services platforms.
TechBio in the actual biology and strain development will continue to be impactful, but business models may be challenged as nascent tool and platform companies struggle to capture sufficient value in the industrial biotech value chain.
Hello! My name’s Matt Kirshner, and I’m an investor and incubation lead at Ferment, a venture studio building product-focused companies on top of the world’s most powerful biotechnology platforms. Previously, I worked across multiple roles at Ginkgo Bioworks and as a management consultant in the biotechnology space at Putnam Associates.
This post benefited greatly from the company-building work my colleagues (thank you Jason, Brian, Jess, and Mia!) do at Ferment everyday, as well as the thought-partnership of many of the EIRs and operators we work with everyday (a special shout-out to Richard Pieters and Chris Schroder for all the deep thinking on similar topics they’ve shared over the past few months).
As a disclaimer, companies described in this post may be partners or portfolio companies of Ferment. The views expressed here do not reflect those of anyone but my own. Feel free to send a note to matt@ferment.co to connect.
so so good. thanks for writing!