Isomorphic Labs Secures $600M Funding to Advance AlphaFold 3-Powered Drug Discovery and Expand Pharma Partnerships

The release of AlphaFold 2 has sparked a revolution in the field of modeling protein structures and their interactions, offering broad applications for protein modeling and design.The AlphaFold 3 model, on the other hand, adopts a substantially updated diffusion-based architecture, capable of predicting the joint structures of complexes that include proteins, nucleic acids, small molecules, ions, and modified residues.The new AlphaFold model shows a significant improvement in accuracy compared to many previous specialized tools: it is much more accurate than state-of-the-art docking tools in protein-ligand interaction prediction; it outperforms nucleic acid-specific prediction tools in protein-nucleic acid interaction prediction; and it also greatly enhances the accuracy of antibody-antigen prediction compared to AlphaFold-Multimer v.2.3.

The figure below showcases AlphaFold 3 (AF3) — a model capable of predicting the structures of nearly all types of molecular complexes in the Protein Data Bank (PDB) with high accuracy (Fig. 1a, b). Except for one category, its performance significantly surpasses methods dedicated to specific tasks across all other categories (Fig. 1c), including higher accuracy in predicting protein structures and protein-protein interaction structures.

Figure 1 AF3 Accurately Predicts the Structure of Biomolecular Complexes

Figure 2: Architecture and Training Details

This achievement was realized through significant improvements to the AF2 architecture and training process (Fig. 1d), enabling it to accommodate more general chemical structures while enhancing data learning efficiency. The system reduces multiple sequence alignment (MSA) processing by replacing AF2's evoformer with a simpler pairformer module (Fig. 2a). Additionally, it employs a diffusion module to directly predict raw atomic coordinates, replacing the structural module in AF2 that operates based on amino acid-specific frameworks and side-chain torsion angles (Fig. 2b). The multiscale nature of the diffusion process (low noise levels encourage the network to refine local structures) also allows for the elimination of stereochemical loss and most special handling of bonding patterns in the network, easily adapting to arbitrary chemical compositions.

In the entire process of drug development, a series of advanced AI models (such as ISO models) have become key drivers. The structural prediction of ligand-protein complexes is a crucial component in the drug discovery process because it allows scientists to understand the intricate interaction networks between proteins and small molecules. These interactions partly lead to the inherent functions of these molecules inhibiting or enhancing the proteins they interact with. In this regard, AlphaFold 3, jointly developed by Isomorphic Labs and Google DeepMind, has revolutionized the field, especially for novel protein targets that are poorly documented in the literature. During the target discovery phase, AlphaFold 3 enables scientists to deeply analyze protein structures, thereby effectively predicting their interactions with small molecules. In the compound design stage, generative models create entirely new molecular structures tailored to the characteristics of the target. Drug designers, utilizing an internal suite of models, can not only predict the binding efficacy of molecules to targets but also meticulously investigate and optimize critical parameters such as in vivo solubility, permeability, and metabolic properties of the molecules. This greatly reduces the number of molecules requiring laboratory synthesis and testing, significantly accelerating the identification of candidate drugs.

In the commercial landscape of AI pharmaceuticals, Isomorphic Labs actively expands collaborations, leveraging its technological advantages to move forward hand in hand with industry giants. In January 2024,Isomorphic Labs Partners with Novartis, a Leader in Innovative Drug Development, for Strategic Research Collaboration Focused on Developing Small Molecule Therapies for Three Highly Challenging TargetsIn just over a year, significant progress has been made by both parties. Building on this, in February 2025, Isomorphic Labs and Novartis announced the expansion of their collaboration scope, adding up to three new research projects, further deepening their exploration in innovative drug development.

The strategic cooperation between Isomorphic Labs and Eli Lilly is equally noteworthy.The collaboration focuses on discovering small-molecule therapies targeting multiple undisclosed targets, reflecting the shared commitment of both parties to advancing breakthrough drug design through cutting-edge science. Under the terms of the two agreements, which are highly attractive, there are 17 milestone payments with potential earnings reaching up to $3 billion—equivalent to 1.5 times the total global AI-driven drug discovery funding in 2023—highlighting the immense potential of the collaborative projects.

The ability to monetize technology is a strong support for Isomorphic Labs' commercial collaborations. Its AI engine has achieved significant breakthroughs in the field of solid tumor treatment. The oral small-molecule inhibitor ISOL-003, designed to target the KRAS G12D mutation, demonstrated outstanding performance in preclinical studies. Compared to Amgen's marketed drug AMG 510, ISOL-003 exhibits binding affinity (Kd=0.03nM vs 300nM) that is four orders of magnitude higher and successfully crosses the blood-brain barrier.

Talent and technical facilities provide a solid backing for the continuous leadership of Isomorphic Labs. In 2024, Demis Ha, winner of the Nobel Prize in Chemistry...Under the leadership of ssabis, a team汇集全球顶尖科研人才的团队应运而生，132人的团队中囊Including 9 members of the National Academy of Sciences, 17 corresponding authors of the main journals Nature and Science, through a 1:1 pairing R&D mechanism between algorithm engineers and biologists, the translation cycle from research papers to patents has been shortened to 11 months, increasing efficiency by 7 times compared to the industry average. Meanwhile, the company independently developed the Luminous computing architecture, equipped with an AI supercomputing cluster consisting of 4096 TPU v5 chips, capable of completing molecular dynamics simulations at the level of 1021 FLOPs in a single operation, equivalent to three months' workload of traditional supercomputers. This lays a solid foundation for complex drug development simulations and analysis, ensuring that Isomorphic Labs steadily advances in the AI pharmaceuticals sector.

03 Future Outlook

In terms of pipeline layout, Isomorphic Labs has already shown initial success, building a product matrix covering 11 preclinical projects, widely spanning three key areas: oncology, neurodegenerative diseases, and rare diseases. Among these, the Alzheimer's disease treatment drug ISOL-007, targeting misfolded Tau proteins, has made particularly remarkable progress, with an expected IND submission in Q2 of 2025. The time from target discovery to this stage took only 19 months, demonstrating an impressive speed advantage compared to the traditional drug development cycle, which often takes 5 to 7 years, bringing new hope to many patients suffering from such diseases.

In terms of technological expansion, Isomorphic Labs, having secured sufficient funding, is making bold strides toward new frontiers by delving into the uncharted territory where cryo-electron microscopy analysis meets AI. Its newly launched Helios platform, powered by the 3D-HELMETS algorithm, achieves automated atomic modeling at a resolution of 4.2Å, with an efficiency up to 120 times higher than traditional manual modeling. This breakthrough holds the potential to overcome the long-standing challenge of membrane protein drug design, thereby unlocking the trillion-dollar market for GPCR-targeted drugs and opening up entirely new horizons for pharmaceutical research and development.

However, the road ahead is not smooth. Isomorphic Labs also faces many challenges, such as issues related to data quality and explainability, including noisy biological data and the black box nature of AI models, which need to be addressed. The effectiveness of AI-predicted molecules still requires repeated validation through rigorous experiments and clinical trials, and global pharmaceutical regulatory agencies are still exploring review frameworks suitable for AI-driven drugs.

Author: Zhancai Hou, Changjian Huang

Share

Like

Click to see