Isomorphic Labs builds IsoDDE, the Isomorphic Labs Drug Design Engine, a proprietary computational system designed to move parts of drug discovery from the physical laboratory onto a computer. The platform combines molecular-structure prediction, binding-affinity estimation, druggable-pocket identification, generative molecular design, and multi-objective optimization.

A project starts with drug designers defining a biological target, molecule type, desired mechanism of action, and constraints for potency, selectivity, solubility, toxicity, and manufacturability. Building on the AlphaFold 3 foundation that Isomorphic jointly developed with Google DeepMind, IsoDDE predicts the target protein's three-dimensional structure and molecular context, including protein-protein interactions, DNA or RNA complexes, and potential drug-binding sites.

IsoDDE searches the protein for potential binding cavities, including allosteric and cryptic pockets that open only when a suitable molecule is present. Rather than screen a fixed library of existing compounds, it generates novel molecular structures against the specification. For each candidate, the system estimates binding affinity, selectivity, solubility, permeability, and other drug-like properties, then iteratively optimizes designs across competing objectives.

Scientists select a small number of high-confidence candidates from the computational run for synthesis and testing in biochemical and cellular assays. The experimental results feed back into the models through a closed loop of prediction, generation, synthesis, testing, and learning. Isomorphic says this process can compress months of physical iteration into two to four days of computational design, followed by targeted laboratory validation.

IsoDDE is used by Isomorphic's internal drug-design teams and partner pharmaceutical research groups rather than offered as self-service software. The system covers small molecules, antibodies, peptides, and molecular glues. Isomorphic reports that it substantially outperforms AlphaFold 3 and the open-source Boltz-2 model on difficult protein-ligand and antibody-antigen prediction tasks.

Isomorphic Labs has two revenue channels: partner-funded discovery programs and wholly owned therapeutic programs. Partners including Lilly, Novartis, and J&J contribute disease biology, experimental assays, and development infrastructure, while Isomorphic provides computational prediction and molecular design. Partnership revenue comprises upfront payments, funded research costs, performance milestones tied to scientific and regulatory progress, and potential downstream royalties on net sales.

The company targets large pharmaceutical companies through bespoke, multi-target research collaborations rather than high-volume software sales. Contracts are technically intensive and slow to negotiate, but each carries substantial economic value. Novartis expanded its collaboration from three to six programs after roughly one year, an example of land-and-expand behavior within this model.

The cost structure has two phases. The computational and discovery phase carries high fixed costs for frontier-model research, AI training infrastructure, specialized scientific labor, and compound synthesis. Once trained, however, models can be reused across multiple targets, modalities, and partners. If Isomorphic advances wholly owned assets into clinical development, its costs increasingly resemble those of a conventional biotech, including IND-enabling studies, manufacturing, regulatory work, and clinical trials.

Partnerships fund engine development and provide external validation, while internal programs retain greater ownership of successful assets. The $2.7B in external financing gives Isomorphic capital to maintain both channels and advance internal candidates toward the clinic rather than licensing every discovery early.

Isomorphic Labs competes in a market converging around vertically integrated drug discovery, where differentiation is shifting from model benchmarks to proprietary experimental data, clinical validation, and the ability to capture economics beyond candidate generation.

Recursion Pharmaceuticals, which combined with Exscientia in late 2024, is the closest scaled competitor. The merged company spans automated wet labs, multimodal models, precision chemistry, and clinical development, with a 23-petabyte imaging dataset, more than 10 clinical programs, over $20B in potential partner milestones, and more than $1B in cash. Recursion focuses on mapping causal biology through image-based phenomics and perturbation experiments rather than atomic structure prediction, while its partnership with Tempus adds patient-level oncology data.

Recursion also co-developed the open-source Boltz family of structure-prediction models, which competes with proprietary structural-biology platforms at the infrastructure layer. Isomorphic reports substantially stronger results than Boltz-2 on antibody-antigen and binding-affinity benchmarks, but the comparisons require prospective validation.

Insilico Medicine has more commercial and clinical validation. Its Pharma.AI platform has produced more than 40 programs, 31 nominated preclinical candidates, 13 IND-cleared programs, and a lead fibrosis drug in Phase III in China. Insilico collaborates with 13 of the 20 largest global pharmaceutical companies and monetizes through software licensing and asset out-licensing.

Isomorphic has substantially greater capital and arguably stronger structural-biology foundations, but Insilico's development speed, commercial flexibility, and clinical footprint provide a benchmark for translating AI claims into pharmaceutical outcomes. As of September 2026, Isomorphic had not publicly named a clinical candidate or provided a clinical-trial timeline.

Schrödinger is an incumbent in physics-based molecular simulation, with more than 30 years of method development, established pharmaceutical workflow integration, and partnered and internal therapeutic programs, including a Phase I asset. Its 2026 launch of Bunsen, an agentic drug-discovery layer deployed with Bristol Myers Squibb, gives pharmaceutical scientists AI capabilities over familiar, validated tools, offering a lower-friction adoption path than a collaboration that shares downstream economics.

In biologics, Generate:Biomedicines has five clinical or clinic-ready candidates, including a lead anti-TSLP antibody in Phase III, while Chai Discovery raised $400M in 2026 and signed collaborations with Novartis, argenx, and Bristol Myers Squibb for de novo antibody design. Novartis works simultaneously with both Isomorphic and Chai, an example of a multi-homing dynamic in which large pharmaceutical companies benchmark multiple AI vendors on overlapping problems and internalize the best techniques.

Lila Sciences, which is developing a broader scientific superintelligence platform, and Google DeepMind itself, whose AlphaFold and Co-Scientist capabilities provide adjacent infrastructure, represent other competitive vectors. Benchling controls research workflows and proprietary biotech data across roughly 1,200 customers, which could make it a distribution layer for third-party scientific models rather than a direct drug-design competitor.

Isomorphic Labs is expanding from small-molecule discovery into a multi-modality, vertically integrated drug-development platform, with each capability increasing its addressable share of global pharmaceutical R&D spending.

IsoDDE began with small-molecule design for the Lilly and Novartis collaborations but has broadened into antibodies, peptides, and molecular glues. The J&J collaboration is its first commercial agreement to span multiple modalities, indicating that the same underlying structural and interaction models can support different drug classes.

The top 10 antibody products generate tens of billions in annual sales. If IsoDDE's reported improvements in antibody-antigen prediction translate prospectively, the platform could support de novo antibody design, affinity maturation, bispecifics, and antibody-drug conjugate components, each with distinct commercial economics.

Isomorphic's transition from discovery partnerships to owning and developing drug candidates offers its largest value-capture opportunity. Partnered programs generate upfront payments and milestones but cede meaningful downstream economics, while wholly owned programs that reach clinical proof of concept and approval would capture more value per asset.

The $2.1B Series B, appointment of a chief medical officer, and establishment of a Cambridge, Massachusetts presence indicate this shift. Isomorphic is building a model in which it generates data, trains models, designs molecules, validates them experimentally, and carries selected candidates through clinical development, moving from research services toward a technology-enabled pharmaceutical business.

IsoDDE's pocket-identification capabilities could expand the ligandable proteome by identifying cryptic and allosteric binding sites on proteins that have resisted conventional screening. This could create first-in-class program opportunities with fewer competitors and stronger intellectual-property claims.

Isomorphic is building its drug-discovery organization around oncology and immunology, which contain validated but difficult targets and biomarker-defined patient populations. The engine's general architecture could later extend into neuroscience, metabolic disease, infectious disease, and rare genetic disorders. The company has also identified bioresilience and pandemic preparedness as adjacent markets that could support government-funded programs and public-private partnerships.

Clinical translation gap: As of September 2026, Isomorphic Labs has not publicly named a clinical candidate or provided a clinical-trial timeline, leaving its competitive position dependent on computational benchmarks and preclinical validation rather than the human pharmacology, safety, and efficacy data that determine whether AI-designed molecules become approved medicines.

Partner multi-homing: Isomorphic's pharmaceutical partners also work with competitors including Chai Discovery, Schrödinger, Insilico Medicine, and their own internal AI teams, creating a risk that partners absorb methods and talent while Isomorphic retains limited differentiation, faces pricing pressure, or receives only the riskiest targets that internal pharma tools have already failed to crack.

Open-source model compression: The release of open-source structure and affinity prediction models such as Boltz-2 under permissive licenses allows pharmaceutical companies, CROs, and startups to build internal workflows without paying a proprietary model provider, potentially eroding the value of Isomorphic's core prediction capabilities and making its durable moat dependent on proprietary experimental data, medicinal-chemistry judgment, and clinical outcomes that have not yet been demonstrated.