Anthropic aims for AI drug development

The widespread impact of artificial intelligence on drug development, particularly in terms of patient accessibility, remains a distant prospect.

This week, during "The Briefing: AI for Science" event, Anthropic unveiled Claude Science, a novel "AI workbench for scientists." This platform aims to consolidate disparate tools and datasets into a unified environment, capable of generating figures and visuals. Anthropic, a prominent leader known for its popular coding tools and robust AI models, positioned this launch as a significant step towards realizing AI's potential to "dramatically accelerate the pace of scientific discovery and the development of healthcare interventions." The company also highlighted a substantial roster of existing biotech and pharmaceutical clients already leveraging Claude.

Furthermore, Anthropic announced an ambitious move into direct drug development. Eric Kauderer-Abrams, the company's Head of Life Sciences, stated that their efforts would concentrate on identifying treatments for "neglected" diseases.

While numerous AI firms, including OpenAI, Amazon, and Google, have actively sought to engage science and pharmaceutical clients with their respective life sciences tools and platforms, Anthropic's latest initiative marks one of the most explicit public endeavors by a leading frontier AI company to independently develop pharmaceuticals. This strategy places Anthropic in the unique position of providing software to potential competitors in the drug manufacturing sector. The company's entry into this domain aligns it with a broader competitive landscape, encompassing AI-first drug companies such as Insilico and Google DeepMind spinout Isomorphic Labs, alongside various biotech startups and established Big Pharma entities that are either developing or acquiring their own AI capabilities.

Anthropic has, however, offered minimal specific details regarding its aspirations within the drug development arena. At the event, Kauderer-Abrams did not elaborate on the company's intended course of action should it identify any promising drug candidates. Furthermore, Anthropic did not respond to inquiries from The Verge seeking additional information, including which diseases it plans to prioritize initially and whether it would collaborate with other organizations for laboratory work, animal testing, clinical trials, or manufacturing.

Experts consulted by The Verge suggested that the ambiguity surrounding Anthropic's intentions mirrors a more general uncertainty within the burgeoning field of AI drug development itself. "AI drug discovery" is a highly expansive term, as Namshik Han, a professor at the University of Cambridge and co-founder of AI biotech startup CardiaTec, clarified. He emphasized that AI finds application at "every single stage of drug discovery," ranging from the identification and optimization of new compounds to bolstering research, data analysis, clinical trials, and even manufacturing processes. Han predicted that every significant pharmaceutical company will, in some capacity, integrate AI into its operations. Matthew Todd, a professor of drug discovery at University College London, reinforced this perspective, noting AI's pervasive presence across drug discovery and research, describing it as a "catchall phrase" due to its diverse applications.

There is no doubt that AI is fundamentally transforming drug development. Han highlighted various initiatives undertaken by pharmaceutical giants such as AstraZeneca, Novo Nordisk, and GSK, noting that AI can already assist in generating potential drug concepts. This includes proposing novel molecules capable of interacting with specific bodily components, like cell receptors, which are either implicated in a particular disease or serve as targets for existing medications. Todd affirmed AI's immense utility in accelerating research and facilitating the "road testing" of new drug hypotheses. Considering Anthropic's specialization in frontier models, the company would presumably leverage generative AI to explore extensive chemical and biological permutations, thereby aiding researchers in making connections that would otherwise be challenging or time-consuming to uncover. This could lead to suggestions for new drug candidates, the identification of novel disease targets, or even the discovery of new applications for existing drugs.

Nevertheless, the journey from an AI-designed drug concept to patient accessibility remains extensive. Todd asserted that the industry is "a long way off" from achieving regulatory approval for human use of an AI-conceived medication. He further stressed that the drug discovery process would not operate autonomously, necessitating continuous human involvement and oversight. Both Todd and Han pointed to the scarcity of publicly accessible, high-quality experimental data – such as information on how various chemicals interact within the human body – as a potential impediment to drug development efforts. They underscored that even in extensively researched biological domains, significant gaps persist in our fundamental understanding.

Frank von Delft, a professor of structural chemical biology at the University of Oxford and head of protein crystallography at the Oxford Centre for Medicines Discovery, stated unequivocally that AI models "haven’t yet come close to making experiments unnecessary." He emphasized that AI is not equipped to resolve many of the most time-consuming aspects of drug discovery. While enthusiasm for advancing AI models is warranted, drug candidates still demand rigorous real-world testing to ascertain their efficacy, toxicity, and practical attributes, ensuring they can be safely prepared, stored, and administered as medicines. This entire process demands substantial investment in skilled personnel, financial resources, and time, particularly during human clinical trials – a phase where numerous promising drug candidates ultimately fail. Von Delft concluded that if Anthropic intends to develop a drug, it "is going to have to spend a lot on experiments."

It appears Anthropic is prepared to undertake this challenge. Over the past year, the company has been actively recruiting biologists and establishing its own wet laboratories. Currently, several job openings for life sciences roles are listed on its career page. Han confirmed that Anthropic has been "actively recruiting" in this domain, noting that several of his academic peers have been contacted by the company. While refraining from disclosing specific names, Han expressed his belief that Anthropic has successfully attracted talent from both major pharmaceutical companies and esteemed academic institutions.

Considering this multifaceted complexity, regardless of the specific disease Anthropic chooses to address, any significant return on investment is likely many years away – at a minimum, the better part of a decade, reflective of the typical duration required for a new drug to navigate clinical trials. Todd reiterated that there's "always a big lag time" associated with testing medicines, emphasizing, "It takes time to show experimentally that something’s safe." To date, no AI-designed drug has successfully completed clinical trials and secured FDA approval for market release. While some AI-developed candidates have progressed to clinical trials, it remains challenging to ascertain the precise extent of AI's contribution, its specific application within the process, or whether these candidates demonstrably outperform conventionally developed drugs. While AI can accelerate certain aspects of the discovery phase, drugs ultimately must validate their efficacy and safety through the established, rigorous, and time-intensive experimental methods conducted in real-world settings.