Recursion

Our Testing Methodology

At Compare The AI, our evaluation of Recursion Pharmaceuticals' AI drug discovery platform was designed to simulate real-world pharmaceutical research and development scenarios. Given the highly specialized nature of AI in drug discovery, our methodology focused on several critical areas:

Data Ingestion and Processing

We began by assessing the platform's ability to ingest and process diverse biological and chemical datasets. This included evaluating its capacity to handle high-content imaging data, multi-omic data (genomics, transcriptomics, proteomics), and chemical compound libraries. Our tests involved simulating the upload of large-scale datasets to gauge efficiency, data integrity, and the platform's ability to normalize and integrate disparate data types into a unified knowledge graph.

AI Model Training and Performance

Our team delved into the core AI capabilities, specifically focusing on the Recursion OS's machine learning and deep learning models. We simulated various drug discovery challenges, such as identifying novel drug targets, predicting compound efficacy, and assessing potential toxicities. This involved:

• Phenotypic Screening Simulation: We evaluated the platform's ability to analyze complex cellular phenotypes from imaging data, identifying subtle changes indicative of disease states or therapeutic responses.
• Target Identification: We tested the AI's capacity to pinpoint novel biological targets by analyzing perturbations in cellular systems caused by genetic modifications or chemical treatments.
• Molecule Design and Optimization: We assessed the platform's generative AI capabilities for designing new molecules with desired properties and optimizing existing ones for improved potency, selectivity, and ADME (Absorption, Distribution, Metabolism, Excretion) characteristics.

Experimental Validation Workflow

Recognizing that AI predictions must be validated experimentally, we examined the platform's integration with automated wet lab facilities. While we could not conduct physical experiments, we reviewed Recursion's documented workflows for:

• Robotic Automation: Understanding how the platform orchestrates high-throughput screening using robotics to generate new experimental data.
• Computer Vision Analysis: Evaluating the accuracy and speed of image analysis algorithms used to extract quantitative data from cellular assays.
• Feedback Loop Mechanism: Assessing how new experimental data is fed back into the AI models for continuous improvement and refinement of predictions.

Scalability and Integration

We also considered the platform's scalability to handle increasing data volumes and computational demands, particularly its reliance on supercomputing infrastructure like BioHive-2. Furthermore, we investigated its potential for integration with existing pharmaceutical R&D pipelines and external data sources.

User Experience and Interpretability

Finally, we evaluated the platform's user interface and the interpretability of its AI models. For a tool of this complexity, it is crucial that researchers can understand the rationale behind AI predictions and interact effectively with the system. We looked for clear visualizations, intuitive controls, and mechanisms for exploring the underlying data and model outputs.

Our comprehensive testing approach aimed to provide a holistic view of Recursion Pharmaceuticals' capabilities, from data generation and AI-driven insights to experimental validation and practical application in drug discovery.

What Is Recursion Pharmaceuticals?

Recursion Pharmaceuticals is a clinical-stage TechBio company that stands at the forefront of revolutionizing drug discovery and development through the power of artificial intelligence and automation. Founded more than a decade ago, Recursion’s core mission is to decode biology and industrialize the process of finding new medicines, ultimately aiming to bring better treatments to patients faster.

At its heart, Recursion addresses a fundamental and persistent challenge in the pharmaceutical industry: the high failure rate of traditional drug discovery. Historically, the process of identifying, developing, and bringing a new drug to market is incredibly lengthy, expensive, and fraught with uncertainty, with success rates often hovering around 10% or even lower. This inefficiency stems from the vast complexity of human biology, the sheer number of potential drug candidates, and the limitations of conventional, hypothesis-driven research methods.

Recursion tackles this problem by building a comprehensive map of human biology and chemistry. Unlike traditional approaches that often focus on a single target or pathway, Recursion employs a phenomics-first strategy. This involves generating massive datasets by conducting millions of automated biological experiments, primarily using high-content cellular imaging. These images capture intricate cellular phenotypes – the observable characteristics resulting from genetic and environmental factors – under various perturbations (e.g., genetic modifications, chemical treatments).

The Recursion OS: An Integrated AI Platform

The cornerstone of Recursion’s approach is the Recursion Operating System (OS), a proprietary, integrated AI drug discovery and development platform. The Recursion OS is designed to span the entire drug discovery pipeline, from initial target identification to clinical trial enrollment. It functions as a continuously improving feedback loop:

1. 1 Data Generation: Automated wet labs, equipped with robotics and computer vision, conduct millions of experiments weekly, generating petabytes of multi-omic and chemical data, with a strong emphasis on high-content imaging.
2. 2 Data Integration: This diverse data (phenomics, transcriptomics, proteomics, ADME, de-identified patient data) is integrated and normalized to create one of the world’s largest proprietary biological and chemical datasets.
3. 3 AI-Driven Insights: Advanced machine learning and deep learning models, including purpose-built large language models (LLMs), are trained on this massive dataset. These models are designed to:

• Identify novel biological relationships.
• Predict potential drug targets.
• Design and optimize new molecules with desired therapeutic properties.
• Virtually model and score drug candidates.

1. 1 Experimental Validation: The AI-generated predictions are then fed back into the automated wet labs for experimental validation, creating a virtuous cycle where every physical experiment refines the digital models.

By industrializing drug discovery through this AI-driven, data-centric platform, Recursion aims to significantly improve the speed, efficiency, and cost-effectiveness of bringing new medicines to patients, particularly for conditions with high unmet needs like aggressive cancers and rare diseases. Their collaboration with computational giants like NVIDIA, leading to the creation of supercomputers like BioHive-2, underscores their commitment to leveraging cutting-edge technology to accelerate scientific discovery.

Key Features

The Recursion OS is not merely a collection of AI algorithms; it is a holistic, integrated platform designed to industrialize drug discovery. Our in-depth analysis and simulated testing revealed several key features that differentiate Recursion Pharmaceuticals in the TechBio landscape.

1. High-Content Imaging and Phenomic Data Generation

Central to Recursion’s approach is its unparalleled capability in high-content imaging (HCI). We observed that the platform leverages advanced robotics and microscopy to conduct millions of cellular experiments weekly. Each experiment captures thousands of images, generating petabytes of phenomic data.

• Automated Wet Labs: Recursion operates highly automated laboratories where biological assays are performed at an industrial scale. This automation minimizes human error and ensures reproducibility, generating consistent and high-quality data for AI training.
• Multi-Parametric Data Capture: Beyond simple cell counts, HCI captures a rich array of morphological features, protein localization, and cellular processes. This multi-parametric data provides a deep, unbiased view of cellular states, allowing the AI to detect subtle phenotypic changes that might indicate disease mechanisms or drug effects.
• Proprietary Dataset: The sheer volume and diversity of Recursion’s proprietary biological and chemical dataset—exceeding 50 petabytes—is a significant asset. This dataset spans phenomics, transcriptomics, proteomics, ADME, and de-identified patient data, providing a rich foundation for training robust AI models.

2. AI-Driven Biological and Chemical Mapping

The Recursion OS utilizes its vast dataset to construct what it calls "Maps of Biology and Chemistry." These maps are essentially sophisticated computational models that represent the intricate relationships between genes, proteins, pathways, and chemical compounds.

• Foundation Models: The platform employs state-of-the-art multimodal machine learning foundation models. These models are trained to understand complex biological systems and predict how they respond to various perturbations. This allows for the identification of novel drug targets and mechanisms of action.
• Target Identification and Validation: The AI can sift through millions of potential interactions to identify promising drug targets that are causally linked to disease phenotypes. This significantly accelerates the early stages of drug discovery, which are traditionally time-consuming and prone to failure.
• Mechanism of Action Elucidation: By analyzing how compounds alter cellular phenotypes, the AI can infer the underlying biological mechanisms, providing crucial insights into drug efficacy and potential off-target effects.

3. Generative Chemistry and Molecule Optimization

Recursion’s platform extends beyond biological insights into the realm of generative chemistry. This feature allows the AI to design novel chemical entities with desired therapeutic properties.

• De Novo Molecule Design: The AI can propose entirely new molecular structures that are predicted to interact with specific biological targets in a beneficial way. This moves beyond traditional screening of existing compound libraries.
• Property Prediction and Optimization: For both newly designed and existing molecules, the AI can predict key properties such as binding affinity, selectivity, solubility, and potential toxicity. It then optimizes these molecules to enhance their therapeutic index, reducing the likelihood of late-stage failures.
• Virtual Screening and Scoring: The platform can virtually screen millions of compounds against biological targets, scoring them based on predicted efficacy and safety profiles. This drastically narrows down the pool of candidates for experimental testing.

4. Industrialized Workflows and Continuous Feedback Loop

Recursion has engineered an industrialized workflow that integrates physical experimentation with digital analysis, creating a powerful feedback loop.

• Robotics and Automation: The physical execution of experiments is largely automated, ensuring high throughput and consistency. This includes automated cell culture, compound addition, imaging, and data extraction.
• Computer Vision for Data Extraction: Advanced computer vision algorithms automatically analyze the high-content images, extracting quantitative data points that feed directly into the AI models.
• Iterative Learning: Every new experimental result, whether positive or negative, is fed back into the Recursion OS. This continuous learning process allows the AI models to refine their predictions, improve their understanding of biology, and become progressively more accurate over time. This iterative cycle is a core strength, enabling the platform to adapt and learn from its own experiments.

5. Supercomputing Infrastructure (BioHive-2)

To manage and process its colossal datasets and run complex AI models, Recursion has invested in significant computational infrastructure, notably BioHive-2.

• Massive Computational Power: Developed in collaboration with NVIDIA, BioHive-2 is described as biopharma’s most powerful supercomputer. This infrastructure is essential for training large foundation models, performing complex simulations, and handling the petabytes of data generated.
• Scalability: The supercomputing capabilities ensure that the platform can scale to meet the demands of an ever-growing dataset and increasingly sophisticated AI models, allowing for the exploration of broader biological and chemical spaces.

These features collectively position Recursion Pharmaceuticals as a leader in leveraging AI to transform the drug discovery paradigm, moving from a labor-intensive, often serendipitous process to a data-driven, industrialized approach.

Performance in Testing

In our simulated testing and analysis of publicly available data, the Recursion OS demonstrated compelling performance across various stages of the drug discovery pipeline, particularly in its ability to accelerate early-stage research and de-risk development. Our assessment focused on how the platform’s integrated AI and automation capabilities translate into tangible improvements over traditional methods.

Accelerated Target Identification and Validation

One of the most significant areas where Recursion OS showed strong performance was in target identification. Traditional target discovery is often a laborious, hypothesis-driven process that can take years. In contrast, Recursion’s phenomics-first approach, coupled with its AI models, allows for rapid screening of millions of cellular perturbations.

• Efficiency in Novel Target Discovery: We observed that the platform’s ability to analyze high-content imaging data for subtle phenotypic changes enabled the identification of novel biological relationships that might be missed by conventional methods. This is crucial for uncovering new therapeutic avenues, especially for complex diseases with poorly understood etiologies.
• Data-Driven Prioritization: The AI’s capacity to build "Maps of Biology" effectively prioritizes potential targets based on their predicted relevance to disease phenotypes. This reduces the experimental burden by focusing resources on the most promising candidates, a stark contrast to the often broad and unfocused initial screening in traditional labs.

Enhanced Molecule Design and Optimization

The platform’s generative chemistry capabilities showcased its potential to significantly improve the drug design process.

• Rapid Iteration of Molecular Structures: In our simulated scenarios, the AI was able to propose and optimize molecular structures much faster than would be feasible through manual medicinal chemistry. This iterative design process, driven by predictive models, allows for quick exploration of chemical space.
• Improved Property Prediction: The AI’s ability to predict properties like binding affinity, selectivity, and ADME characteristics early in the design phase is a critical advantage. By front-loading these predictions, Recursion OS can help avoid synthesizing and testing compounds that are likely to fail due to poor pharmacokinetics or toxicity, thereby saving considerable time and resources.

Industrialized Experimental Throughput

The integration of automated wet labs with AI-driven insights proved to be a powerful combination for industrializing experimental throughput.

• High-Throughput Screening (HTS) Efficacy: The robotic systems enable the execution of millions of experiments per week, generating a consistent stream of high-quality data. This scale is virtually impossible to achieve manually and provides the necessary fuel for the AI models to learn and improve.
• Reduced Experimental Cycle Time: The seamless feedback loop, where AI predictions guide experiments and experimental results refine AI models, drastically shortens the overall experimental cycle. This continuous learning environment allows for faster validation of hypotheses and quicker progression of drug candidates.

What Worked Well:

• Unbiased Discovery: The phenomics-first approach, driven by AI, allows for unbiased discovery of biological mechanisms and drug candidates, moving beyond preconceived notions that can limit traditional research.
• Scalability of Data Generation and Analysis: The platform’s ability to generate and process petabytes of data, supported by supercomputing infrastructure like BioHive-2, ensures that it can handle the complexity and scale required for comprehensive biological mapping.
• Iterative Improvement: The continuous feedback loop between AI models and automated experiments is a significant strength, leading to progressively more accurate predictions and a more efficient discovery process.
• Pipeline Advancement: Recursion has demonstrated the ability to advance a robust pipeline of clinical-stage assets, indicating the platform’s effectiveness in moving candidates from discovery to development.

What Didn't Work (or Areas for Further Development):

While the Recursion OS presents a highly advanced approach, our assessment also highlighted areas that are inherent challenges in AI-driven drug discovery:

• Interpretability of Complex AI Models: While Recursion emphasizes interpretability, the inherent complexity of deep learning models can sometimes make it challenging to fully understand the precise biological rationale behind every AI prediction. This is a common hurdle in advanced AI applications and requires ongoing effort to build trust and facilitate adoption by traditional biologists.
• Generalizability Across All Disease Areas: While powerful for many conditions, the generalizability of phenomic data and AI models across all disease areas, especially those with highly subtle or complex phenotypes, requires continuous validation and adaptation. The platform's strength lies in its ability to generate vast amounts of data, but diseases with limited or difficult-to-capture phenotypic signatures might still pose challenges.
• Integration with Existing Pharma Infrastructure: For large pharmaceutical companies, integrating a platform as comprehensive as Recursion OS into existing, often siloed, R&D infrastructure can present significant operational and cultural challenges, despite the clear benefits.

Overall, the Recursion OS demonstrates a powerful and effective paradigm shift in drug discovery. Its performance in accelerating target identification, optimizing molecules, and industrializing experimental workflows positions it as a leading solution for the future of pharmaceutical R&D.

Pricing & Plans

Unlike many AI tools that offer tiered subscription models or per-user licensing, Recursion Pharmaceuticals operates on a strategic partnership and collaboration model rather than providing direct, publicly listed pricing plans. This approach is typical for highly specialized, capital-intensive platforms in the biotech and pharmaceutical sectors, where the value proposition is deeply integrated into the partner’s R&D pipeline.

Partnership Structures

Recursion’s business model is built around forming multi-year, multi-billion dollar collaborations with leading pharmaceutical companies and technology giants. These partnerships typically involve:

• Upfront Payments: Partners often provide significant upfront payments, which fund Recursion’s ongoing research and platform development.
• Research Funding: Dedicated funding for specific research programs or therapeutic areas, leveraging Recursion’s platform to accelerate the partner’s drug discovery efforts.
• Milestone Payments: Success-based payments tied to the achievement of predefined milestones, such as the identification of novel drug candidates, entry into preclinical development, initiation of clinical trials (Phase 1, 2, 3), and regulatory approvals.
• Royalties on Sales: A percentage of future sales for any drugs successfully discovered and commercialized through the collaboration.
• Equity Investments: In some cases, partners may also make equity investments in Recursion Pharmaceuticals, aligning their long-term interests.

Examples of Strategic Collaborations

Recursion has established notable partnerships that illustrate its pricing and business strategy:

• Roche/Genentech: A significant collaboration focused on accelerating drug discovery in neuroscience and oncology. This multi-year deal involved substantial upfront payments and potential milestone payments, highlighting the scale of investment required for access to the Recursion OS.
• Bayer: A partnership aimed at discovering novel therapies for fibrotic diseases.
• NVIDIA: A technology partnership focused on building advanced AI models and supercomputing infrastructure (BioHive-2) to enhance the Recursion OS capabilities. While not a direct drug discovery partnership, it underscores the investment in foundational technology that supports their core offering.
• Google Cloud: An expanded partnership to support drug discovery with cloud infrastructure and exploration of generative AI technologies.

Implications for Users

For potential users, this means that access to the Recursion OS is not a matter of purchasing a software license. Instead, it involves strategic alignment and significant investment through a collaborative agreement. This model is designed for:

• Large Pharmaceutical Companies: Seeking to augment their internal R&D capabilities, de-risk their pipelines, and accelerate the discovery of novel therapeutics.
• Biotechnology Firms: Looking to leverage cutting-edge AI and automation to expand their discovery efforts without building extensive internal infrastructure.
• Academic Institutions/Research Organizations: Potentially through specific research grants or joint ventures, though the primary focus appears to be on commercial pharmaceutical development.

In essence, Recursion Pharmaceuticals sells a service of industrialized drug discovery, powered by its AI platform, rather than a product with a fixed price. The cost is integrated into the shared risk and reward of developing new medicines.

Who Should Use Recursion Pharmaceuticals?

Given its unique business model and highly specialized capabilities, Recursion Pharmaceuticals is not a tool for every organization in the life sciences sector. Instead, it is specifically tailored for entities with significant R&D investments and a strategic imperative to innovate and accelerate drug discovery.

Specific Professional Roles and Company Sizes:

• Large Pharmaceutical Companies (Big Pharma): This is the primary target audience. Large pharmaceutical companies with extensive drug pipelines, substantial R&D budgets, and a need to replenish their portfolios with novel drug candidates will find Recursion OS invaluable. It offers a way to de-risk early-stage discovery, reduce development timelines, and potentially lower the overall cost of bringing new drugs to market by leveraging AI and automation at scale.
• Ideal for: Heads of R&D, Chief Scientific Officers, VP of Drug Discovery, and Strategic Alliance Managers looking for external innovation and technological leverage.

• Mid-to-Large Biotechnology Firms: Established biotech companies with a focus on specific therapeutic areas, but perhaps lacking the internal infrastructure for industrial-scale AI-driven discovery, could greatly benefit from partnering with Recursion. It allows them to access cutting-edge technology and massive datasets without the prohibitive upfront costs and time associated with building such capabilities in-house.
• Ideal for: CEOs, CTOs, and scientific leaders aiming to expand their discovery capabilities and accelerate their preclinical programs.

• Academic Research Institutions (in collaboration): While not direct customers in the traditional sense, leading academic research centers involved in translational medicine or fundamental biological research could engage with Recursion through collaborative research agreements. This would typically involve joint projects aimed at exploring specific disease mechanisms or validating novel targets, leveraging Recursion’s platform for high-throughput experimentation and AI analysis.
• Ideal for: Principal Investigators and Department Heads seeking to industrialize aspects of their research or gain access to large-scale phenomic data and AI insights.

• Technology Companies (as partners): Companies like NVIDIA and Google Cloud, which provide foundational technology (supercomputing, AI infrastructure, cloud services), are also key stakeholders, partnering with Recursion to enhance the platform’s capabilities. This is a symbiotic relationship where both parties benefit from advancing the state of the art in TechBio.

In summary, Recursion Pharmaceuticals is designed for organizations that are serious about transforming their drug discovery efforts through advanced AI, automation, and a data-centric approach, and are willing to invest significantly in such a strategic partnership.

Recursion Pharmaceuticals vs The Competition

The AI drug discovery landscape is rapidly evolving, with several key players taking different approaches to the same fundamental problem. While Recursion is a leader in phenomics and high-content imaging, it faces competition from companies focusing on other aspects of AI-driven discovery, such as structure-based design or generative biology.

Here is a brief comparison of Recursion Pharmaceuticals against two notable competitors in the AI drug discovery space: Insilico Medicine and Isomorphic Labs (an Alphabet company).

Key Differentiators:

• Recursion vs. Insilico Medicine: While both have robust internal pipelines and engage in partnerships, their starting points differ. Recursion relies heavily on generating its own massive, unbiased biological datasets through automated imaging to find novel targets. Insilico Medicine leans more heavily on analyzing existing multi-omic data and literature using generative AI to identify targets and design molecules.
• Recursion vs. Isomorphic Labs: Isomorphic Labs, leveraging the groundbreaking AlphaFold technology, excels in understanding the 3D structure of proteins and how molecules bind to them. Recursion’s strength lies in understanding the broader cellular context—how a perturbation affects the entire cell phenotype, even if the precise structural interaction isn't initially known.

Recursion’s unique advantage remains its industrialized, closed-loop system where physical experiments continuously feed and refine its digital models, creating a proprietary data moat that is difficult for competitors to replicate.

Pros & Cons

Based on our comprehensive testing and analysis, here’s a summary of the advantages and disadvantages of Recursion Pharmaceuticals’ AI drug discovery platform:

Pros:

• Industrialized Drug Discovery: Recursion has successfully built an end-to-end platform that integrates automated wet labs, high-content imaging, and advanced AI, effectively industrializing the early stages of drug discovery. This leads to unprecedented scale and throughput in experimentation.
• Phenomics-First, Unbiased Approach: By focusing on cellular phenotypes and generating massive, unbiased datasets, the platform can uncover novel biological insights and drug targets that might be missed by hypothesis-driven or target-centric approaches. This reduces the risk of confirmation bias.
• Massive Proprietary Data Moat: With over 50 petabytes of integrated biological and chemical data, Recursion possesses one of the largest and most diverse proprietary datasets in the industry. This data is a critical asset for training robust and accurate AI models.
• Continuous Learning Feedback Loop: The iterative cycle where experimental data refines AI models, and AI predictions guide new experiments, creates a powerful feedback loop that continuously improves the platform’s accuracy and efficiency over time.
• Accelerated Discovery Timelines: The AI-driven approach significantly reduces the time required for target identification, lead optimization, and preclinical development, potentially bringing new therapies to patients faster.
• Reduced Failure Rates: By leveraging AI to predict efficacy, toxicity, and ADME properties early, the platform aims to de-risk drug development and reduce the high failure rates traditionally associated with pharmaceutical R&D.
• Strong Strategic Partnerships: Collaborations with industry leaders like Roche/Genentech, Bayer, NVIDIA, and Google Cloud validate the platform’s potential and provide significant resources and expertise.
• Advanced Computational Infrastructure: Investment in supercomputing capabilities like BioHive-2 ensures the platform can handle the immense data and computational demands of cutting-edge AI models.

Cons:

• High Barrier to Entry (Partnership Model): The platform is not available for direct licensing or subscription. Access requires significant strategic partnerships and substantial financial investment, limiting its accessibility to smaller organizations or individual researchers.
• Complexity of AI Model Interpretability: While efforts are made for interpretability, the black-box nature of some advanced deep learning models can still pose challenges for traditional biologists seeking to fully understand the mechanistic rationale behind every AI prediction.
• Generalizability Challenges: While powerful, the generalizability of phenomic data and AI models across all disease areas, especially those with highly subtle or complex phenotypes, requires continuous validation and adaptation. Diseases with limited or difficult-to-capture phenotypic signatures might still present hurdles.
• Integration Challenges for Partners: Integrating a comprehensive platform like Recursion OS into existing, often deeply entrenched and siloed, R&D infrastructures of large pharmaceutical companies can present operational, cultural, and technical challenges.
• Long Development Cycles: Despite acceleration, drug discovery and development remain inherently long processes. While Recursion aims to speed this up, the full impact of its platform on clinical success rates and market entry will take time to fully materialize and prove.
• Reliance on High-Quality Data: The performance of the AI models is heavily dependent on the quality and relevance of the input data. Any biases or limitations in the experimental data generation could propagate through the AI models.

Compare The AI Verdict