You Have No Moat – The Great AI Democratization Spells New Challenges and Opportunities for SynBio Companies

The following article is an opinion piece written by Gideon Lapidoth. The views and opinions expressed in this article are those of the author and do not necessarily reflect the official position of Technology Networks.

The release of AlphaFold marked a watershed moment in the realm of computational protein design, ushering in a new era of possibilities. For years, the field had been dominated by the Rosetta software package, arguably the most successful protein modeling and design software ever developed. However, comprising multiple algorithms and decades of human experience, it requires specialist expertise to utilize effectively. Now, with the democratization of computational design facilitated by new AI tools, such as AlphaFold, ESM, ProGen and ProteinMPNN, synthetic biology companies must shift focus if they are to remain competitive.

The paradigm of companies being able to merely offer computational design as a service is over.

The old order – computational design as a service

Traditionally, those looking to apply machine learning and optimization algorithms had to handcraft specific sets of features that would be used to evaluate the quality of the output. For example, when developing a program to infer a person's emotion from an image, domain experts would curate handpicked features to assess the range of facial expressions, such as the curvature of the eyebrows and the shape of the mouth. This process required meticulous feature selection and tuning, and a deep understanding of both the problem space and the algorithms involved.

Similarly, with protein design and modeling, the Rosetta suite comprises multiple tunable parameters that must be calibrated according to the desired objective. Therefore, despite the obvious benefits that Rosetta brought to protein design – from improved protein stability to novel structure and functionality – the level of expertise needed to use it successfully restricted its usability to a small number of “Rosetta experts.” This meant that any protein development initiative required outsourcing of the computational design elements, resulting in the emergence of dedicated providers offering computational design as a service.

The democratization of neural networks

The emergence of neural networks, heralded by the release of AlphaFold, transformed the landscape of protein design. Trained neural networks generally do not require the complex handcrafted features of their predecessors. As an example, AlphaFold can take the amino acid sequence of a protein and, with minimal human intervention, produce a structural model faster and more accurately than previous state-of-the-art algorithms. This has opened up computational design to a broader range of users, empowering scientists and researchers who previously lacked the necessary expertise.

In conjunction with the proliferation of AI tools is the adoption of open-source community standards, in which source code is made freely available for anyone to use and improve as desired. This open-source mentality does not stem from purely altruistic intentions, but the necessity to provide a competitive alternative to the tech giants, as revealed in the leaked Google memo admitting that “open-source models are faster, more customizable, more private, and pound-for-pound more capable” than its own models.

The open-source community is taking big tech head-on in the development of new deep-learning models and, in terms of both performance and scale, it appears to be winning. As soon as a new model is announced, there is an immediate scramble to develop comparable, or even superior, alternatives.

Such is the risk to big tech’s supremacy over AI that DeepMind initially did not disclose the source code for AlphaFold, providing only a high-level description of the algorithm and inference results. Yet, even this was enough for coders to quickly develop their own versions, resulting in programs such as RoseTTAFold and OpenFold that can perform on par (and sometimes better) than AlphaFold. The result, as stated in OpenFold’s mission statement is the emergence of an open-source marketplace making available to everyone tools that are “competitive with the performance of state-of-the-art models.”

The implications of an open-source marketplace

The implications of this new world order for the protein development industry are significant. With an abundance of new AI tools making it easier than ever before to develop novel proteins, the traditional business model of providing computational protein design as a service is becoming increasingly unviable.

As ever more users gain access to algorithms requiring lower levels of proficiency, the value-add of pure service companies diminishes. The justification for sharing intellectual property ownership between the contractor and these service providers is similarly eroded, thus destroying the traditional way that such companies previously maximized their value.

Product is king

In this new era, SynBio companies must shift their emphasis from the “how” to the “what” and transition from providing computational design as a service to creating actual products that provide tangible value. This will also address the IP moat problem, as a company’s competitive edge shifts from mere design to building novel proteins and their applications. This will be true for both therapeutics, where a protein’s sequence space and designation can be protected, as well as biomanufacturing in which novel enzymes or engineered organisms can be applied to manufacturing specific molecules. The specific AI model employed to develop these novel enzymes becomes less significant in comparison.

Future success lies beyond design

SynBio companies must therefore recognize and respond to the shifting paradigm if they are to remain competitive. In the open-source marketplace, offering computational design as a service alone is no longer sufficient. They must leverage AI tools and the democratization of computational design to create valuable products that address the needs of a diverse range of industries. By doing so, these companies can establish a stronger market position and stay ahead of the curve by maximizing their value in the evolving landscape of computational design.

About the author: 

Gideon Lapidoth is CEO and co-founder of Enzymit, a bioproduction platform company developing cell-free enzymatic manufacturing technology. He is an expert in developing advanced computational algorithms for the design and creation of novel antibodies and enzymes with superior accuracy and activity. Gideon holds a Master of Science in Biochemistry and Molecular Dynamics from Tel Aviv University and a PhD in Computational Biology from the Weizmann Institute of Science.