TL;DR:
- Former Meta researchers established AI biotech startup EvolutionaryScale, securing $40 million funding from Lux Capital.
- Led by Alexander Rives, the team developed a transformers-based model trained on protein molecule data to predict unknown protein structures.
- EvolutionaryScale’s database holds 700 million potential 3D structures, promising breakthroughs in drug development and pollution cleanup.
- Lux Capital leads the funding round, valuing the startup at $200 million; AI investors Nat Friedman and Daniel Gross participate.
- The AI model accelerates protein structure predictions 60 times faster than AlphaFold, though with slightly less accuracy.
- The fusion of AI and biology is yet to yield a revolutionary commercial boom, contrasting with AI’s influence in other domains.
- EvolutionaryScale joins a series of AI-focused startups securing significant investments, betting on future returns.
- The AI-driven protein folding field requires substantial investment; collaborations with existing players like Isomorphic Labs and Recursion are crucial.
- EvolutionaryScale projects substantial expenditures over three years, acknowledging the long-term timeline for AI-driven biotech advancements.
Main AI News:
In a dynamic leap forward, former researchers from the renowned Meta have orchestrated an impressive feat by establishing a groundbreaking AI biotech startup, EvolutionaryScale, and successfully amassing a substantial investment of at least $40 million. Spearheaded by Alexander Rives, previously at the helm of Meta AI’s avant-garde protein-folding unit until the venture’s cessation in April, this new venture embarks on a transformative journey at the intersection of artificial intelligence and biotechnology.
This innovative team of eight, all hailing from the same pioneering unit, has forged a pioneering path. Drawing inspiration from the likes of OpenAI’s heralded GPT-4 and Google’s Bard, they have ingeniously devised a transformers-based model with an AI language that centers on biology. Their creation stands poised to revolutionize the understanding of protein molecules through comprehensive data analysis, enabling predictions of the enigmatic structures of heretofore unknown proteins. The AI model has been meticulously trained using an expansive dataset focused on protein molecules, resulting in a formidable database housing an impressive 700 million potential three-dimensional structures. These structures hold the key to unlocking a treasure trove of solutions, from groundbreaking drug development for various ailments to eco-friendly approaches for tackling pollution and the production of industrial chemicals.
By the time June had graced the calendar, EvolutionaryScale had set its sights on garnering the support of venture capitalists, aiming to secure the essential seed financing. Their aim was clear: to exponentially amplify the scale of their AI model and, in turn, propel their research endeavors to new heights. In an exclusive look at a pitch document procured by Forbes, it was evident that the startup was diligently striving to enlarge its horizons. It comes as no surprise that the illustrious Lux Capital spearheaded the momentous funding round, which is estimated to have amassed approximately $40 million. Insights from those privy to the deal suggest that the financial influx has placed EvolutionaryScale’s valuation at an impressive $200 million. Furthermore, notable AI aficionados, Nat Friedman and Daniel Gross, were reported to be among the esteemed participants in this groundbreaking venture.
Regrettably, Rives, the trailblazing mind behind EvolutionaryScale, refrained from providing a statement at this juncture. Similarly, Lux Capital, Friedman, and Gross remained reticent in response to the requests for their valuable insights.
Central to this paradigm-shifting endeavor is the complex realm of proteins – intricate molecules intricately woven from chains of amino acids – that serve as the fundamental building blocks for an array of biological entities, spanning bacteria and human cells alike. Their unique functionality is intimately intertwined with their distinctive shapes, a configuration that can undergo alterations upon interactions with various chemicals or other proteins within the biological framework. These shape-shifting traits play a pivotal role in the development of targeted drugs, honing in on specific regions of proteins to address diseases. However, the task of predicting these protean structures is a labyrinthine puzzle, governed by the intricate interplay between myriad atoms within. To exemplify, the interlocking dance of sulfur atoms within a specific amino acid gives rise to the manifestation of curly hair. The implications are profound; this predictive capacity empowers scientists to decode protein functionalities, subsequently steering the course of drug design with a keen focus on three-dimensional configurations.
A noteworthy milestone arrived in 2020 when Google’s subsidiary, DeepMind, astounded the scientific community with the unveiling of AlphaFold – an AI-powered system with the remarkable capability to forecast protein structures. Nobel laureate Venki Ramakrishnan lauded this landmark achievement, heralding it as a “stunning advance” poised to overhaul biological research. While this breakthrough undoubtedly advanced the field, the intricate interplay between proteins and potential drugs remains a formidable challenge, as elucidating the intricacies of these interactions poses a monumental hurdle.
In a significant stride, Rives’ team took center stage with a compelling publication in the esteemed journal Science, underscoring their model’s remarkable feat of accelerating predictions by a factor of 60 compared to AlphaFold. Notably, these predictions exhibit a nuanced trade-off between speed and accuracy. Nevertheless, AI’s integration into drug development has primarily yielded incremental enhancements to efficiency, and the watershed moment characterized by a transformative surge in biological research akin to its text-based counterpart has yet to materialize. Skepticism lingers among traditional pharmaceutical entities, casting doubt upon the eventual convergence of AI and biology into a harmonious symphony of innovation.
The disbandment of Rives’ team from Meta in April occurred against the backdrop of the ChatGPT-fueled rush to harness the commercial potential of generative AI. While Meta redirected its AI efforts toward commercially viable projects, the journey of AI-driven innovation has led to selective pruning. This trajectory finds resonance in OpenAI’s decision to dissolve its robotics division in 2021. Notably, the dissolution of Meta’s protein-focused unit, as initially reported by The Financial Times, transpired alongside broader reconfigurations and strategic realignments to concentrate on profit-generating initiatives, including a diverse array of AI-powered chatbots. In this unfolding narrative, the allure of AI for biology’s imminent commercial returns is somewhat subdued. Established players like Schrödinger, boasting a market capitalization shy of $3 billion, continue to anchor their products in traditional molecular modeling methodologies.
Notably, EvolutionaryScale stands as the latest torchbearer, orchestrating a capital coup in the realm of transformer-based AI research. Among the ranks of unicorn model developers, we encounter remarkable examples such as Inflection AI, reaping a staggering $1.3 billion in June, and Cohere, which made waves with a remarkable $270 million declaration in May. Adept, with its recent $350 million infusion in March, echoes this resounding trend. The fervor extends to the very infrastructure of AI, as highlighted by the substantial $235 million investment disclosed by the much-discussed Hugging Face, valuing the enterprise at an impressive $4.5 billion. It is crucial to acknowledge that many of these monumental investments are predicated on future revenue streams that are projected to burgeon over time. In the same vein, challenges abound; Stability AI, despite securing a sizable $100 million at a valuation of $1 billion the previous year, grapples with revenue generation challenges.
Anticipating further strides in the AI-driven revolution of protein folding necessitates a formidable commitment to investment. DeepMind’s recent foray into drug discovery, marked by the establishment of Isomorphic Labs in December 2022, underscores this point. Meanwhile, Insitro and NASDAQ-listed Recursion stand as noteworthy rivals, boasting over $1 billion in combined support from private and public investors. In this expansive landscape, EvolutionaryScale assumes a pivotal role, positioning itself as a potential partner capable of empowering companies like Insitro and Recursion through its visionary models. However, the road ahead remains arduous; the average timeline for shepherding a drug from its discovery to FDA endorsement spans a formidable 7 to 10 years.
Rives and his visionary team harbor a pragmatic recognition of the moonshot nature of their endeavor. EvolutionaryScale’s ambitious roadmap anticipates a substantial expenditure of $38 million in its inaugural year, with a noteworthy allocation of $16 million towards computing power. As the journey unfolds, costs burgeon, reaching $161 million by the second year and an impressive $278 million by the third. Within this intricate calculus, compute expenses amount to $100 million and $200 million, respectively. Throughout the strategic blueprint, a consistent theme reverberates: the potential of biology AI models to yield tangible products and therapies is an endeavor that demands time, spanning a potentially protracted decade.
Conclusion:
EvolutionaryScale’s groundbreaking approach and substantial funding signify a pivotal step in the convergence of AI and biotechnology. While the transformative potential of AI in biology is yet to fully materialize commercially, startups like EvolutionaryScale are shaping the landscape for future innovations, underlining the need for sustained investments and collaborations to realize the promised breakthroughs.
