[Edaily Reporter Kim Seung-kwon ] Isomorphic Labs, a spin-off from Google DeepMind, is reportedly on the verge of entering the first human clinical trial for a new cancer drug designed using artificial intelligence (AI). If successful, this would mark the first instance of a Big Tech company directly serving as the sponsor of a clinical trial for a drug it designed itself.

Until now, big tech companies such as Microsoft have limited themselves to supporting roles, providing computing infrastructure or AI models to pharmaceutical companies. However, industry experts believe that if Google proceeds with its own clinical trials, the focus of AI-driven drug development will take a significant leap forward—shifting from “technology demonstrations” to “clinical validation.”

An official from the domestic AI-driven drug discovery industry explained, “This will now become a battle for actual human clinical data,” adding, “AI-driven drug discovery is shifting from a competition over ‘who can build a smarter model’ to a competition over ‘who can prove real clinical results first.’”

After AlphaFold Comes “Clinical Trials”… Isomorphic Pursues Vertical Integration from Discovery to Development

Isomorphic Labs’ starting point is the protein structure prediction AI “AlphaFold.” Released by Google DeepMind in 2020, AlphaFold was hailed for solving a long-standing challenge in structural biology by predicting the three-dimensional structure of proteins based solely on amino acid sequences; for this achievement, Demis Hassabis, founder and CEO of Isomorphic Labs, was a co-recipient of the Nobel Prize in Chemistry. Isomorphic Labs spun off from DeepMind as an independent organization with the goal of extending this technology to the entire drug discovery process.

IsoMorphic Labs’ ultimate vision is clear. Considering that traditional drug discovery incurs costs in the billions of dollars yet achieves a success rate of only 10%, IsoMorphic’s goal is not merely to improve speed but to fundamentally raise the probability of success itself.

Conventional drug discovery has centered on proteins with already-known binding pockets where drugs can attach. In contrast, IsoDDE employs a method that infers potential binding pockets based solely on a protein’s amino acid sequence, without requiring ligand information.

The company explained that for cereblon—one of the key targets for proteolysis-induced therapeutics—it predicted not only the existing thalidomide binding pocket but also a cryptic pocket that remains hidden in the absence of a ligand.

The approach to identifying candidate compounds also differs. While the traditional method involves identifying active compounds through high-throughput screening of libraries containing hundreds of thousands to millions of compounds, AI-driven design proposes and prioritizes new chemical structures based on the target’s structure and predicted interactions. Isomorphic Labs announced that IsoDDE achieved higher accuracy than AlphaFold3 in the “Runs N’ Poses” benchmark, which evaluates highly novel protein-ligand combinations.

Colin Murdoch, former president of Isomorphic Labs, stated in a recent interview, “We have begun assembling a dedicated team for human clinical trials, and clinical trials will begin soon,” adding, “Right this very moment, employees at our London office are using AI to design new cancer treatments.”

These claims are backed by technical achievements. According to the technical report on “IsoDDE (Isomorphic Labs Drug Design Engine),” the company’s proprietary drug design engine released last February, IsoDDE more than doubled its accuracy compared to AlphaFold3 for novel target-ligand combinations with low similarity to the training data. It also achieved a 70% success rate in modeling the CDR-H3 loop—a key binding site for antibodies—surpassing AlphaFold3 (58%). The company explained that in binding affinity predictions, it achieved higher accuracy than the industry gold standard, physics-based simulation (FEP+), in a significantly shorter amount of time.

Building on these achievements, Isomorphic Labs secured a $2.1 billion (approximately 2.9 trillion won) Series B funding round last May. The round was led by Thrive Capital, with new participants including MGX, Temasek, and Capital G, joining existing investors such as Alphabet and GV. The industry views Isomorphic Labs—which, just five years after its founding, has partnered with the three major global pharmaceutical companies—Novartis, Lilly, and J&J—and secured capital in the trillions of won—as the AI-driven drug discovery company closest to the actual clinical stage.

Murdoch said, “I hope that someday, once a disease is identified, we’ll be able to design a drug that targets that disease with the push of a button,” adding, “All of this will be made possible by AI tools.”

A Three-Way Race Between OpenAI, Google, and Anthropic… AI-Driven Drug Discovery: The Race to Accelerate Clinical Trials Is Imminent

Isomorphic Labs’ move symbolizes more than just a challenge from Google alone. With OpenAI and Anthropic—the leaders in the generative AI market—recently joining the competition in life sciences and new drug development in earnest, industry observers are assessing that “a platform competition has begun that goes beyond simple research assistance to revolutionize the entire new drug development process.”

The most recent entrant into this race is Anthropic. The company announced the launch of “Claude Science,” a platform dedicated to scientific research, and simultaneously revealed plans to pursue its own drug discovery program targeting neglected diseases—conditions that have received insufficient research due to their low market potential. Claude Science is a research platform that integrates the databases, code execution environments, and analytical tools used by researchers into a single workspace, supporting everything from genomic analysis to protein analysis and compound design.

Particularly noteworthy are the achievements in protein design. Anthropic announced that its AI model, Claude, successfully designed new proteins that bind to specific targets by selecting and applying various AI tools for protein design as needed. Among 1,320 protein candidates capable of actual synthesis and analysis, 354 bound to their targets, resulting in a hit rate of 26.8%. When considering only the candidates that Claude ranked as the top priority in each experiment, the hit rate rises to 49%. Given that the hit rate for conventional drug discovery methods is typically less than 1%, this represents a significant gap.

An Anthropic spokesperson stated, “This demonstrates the potential for AI to accelerate scientific discovery by reducing the expertise, costs, and time required for new drug development.”

OpenAI and Google’s strategies take a different approach. OpenAI unveiled “GPT-Rosalind,” an inference model specialized for the life sciences, last April, and recently launched “GeneBench-Pro,” an evaluation system designed to validate AI’s performance in life sciences research. This is interpreted as a strategy to not only build models but also to establish the very standards for evaluating research performance. Companies such as Amgen, Moderna, and Thermo Fisher are already incorporating GPT-Rosalind into their research workflows.

In contrast, Google has opted for a vertically integrated structure—conducting everything from discovery through preclinical and clinical stages within a single organization—rather than supplying an AI platform via Isomorphic Labs. As the strategies of these three companies diverge, industry observers are summarizing the landscape as follows: “OpenAI focuses on evaluation models, Google on in-house development, and Anthropic on a hybrid model that combines a platform with its own pipeline.”

An official from Hitz, a South Korean AI-based new drug development platform company, stated, “AI competition has now evolved beyond simply creating superior models toward building an ecosystem that encompasses research data, platforms, and regulatory compliance capabilities,” adding, “Ultimately, the market winner is likely to be the company that demonstrates meaningful results in actual clinical trials based on its platform.”