Which platforms run the design-make-test-analyze loop for protein engineering?

If by “design–make–test–analyze” (DMTA/DBTL) you mean a platform that actually closes the loop between computational protein design, physical experimentation, and learning from assay results, the landscape is fairly concentrated.

The important distinction

There are really three categories:

1. Software-first protein engineering

• Cradle
• ProteusAI
• ProGene

These handle the design → data → model → next design loop, but the physical make/test step generally occurs in the customer's lab or through a partner. 2. Foundry / automation platforms

• Ginkgo's Foundry is the clearest example.

Here the differentiator is not necessarily the best protein-design model; it's the ability to physically execute large numbers of designs and return standardized experimental data. Cradle + Ginkgo is an explicit example of coupling AI protein design to automated testing and feeding the results back into the model. Cradle 3. Fully integrated protein-engineering companies

• Generate:Biomedicines
• DPTX
• HemeAI
• Arzeda

These are attempting to make the entire loop one system, rather than selling only the design software. Generate, for example, explicitly describes its platform as generate → build → measure → learn, with experimentation continuously training/refining the system.

One particularly interesting entrant is DPTX, because its stated architecture is unusually literal: automated expression → purification → stability/binding/developability assays → ML-ready data → next design.

If you're asking because you're mapping the competitive landscape for a new protein-engineering platform, I can also break these down by (a) AI design capability, (b) wet-lab automation, (c) assay breadth, (d) degree of autonomy, (e) customer-vs-internal use, and (f) therapeutic vs industrial focus.

The Design-Make-Test-Analyze (DMTA) —or Design-Build-Test-Learn (DBTL) —loop for protein engineering has shifted from manual, piecemeal lab work to fully automated, AI-driven "lab-in-the-loop" platforms . These systems integrate generative artificial intelligence, large language models (LLMs), robotic biofoundries, and automated analytical assays.

The primary systems, architectures, and platforms executing this closed-loop cycle span academic/open-source pipelines, biofoundry integrations, and commercial enterprise solutions:

1. Fully Autonomous Academic & Open-Loop-to-Closed Platforms

• iBioFAB (Illinois Biological Foundry for Advanced Biomanufacturing) / Huimin Zhao Lab Platform: Highlighted in recent autonomous engine frameworks (such as AI-powered platforms published in Nature Communications ), this architecture integrates deep learning and LLMs with robotic hardware to handle gene synthesis, cloning, expression, and high-throughput assays autonomously. It runs multiple iterative cycles with minimal human intervention.
• AssemblyTron & PlasmidMaker: Open-source or specialized automated pipeline architectures built on flexible liquid handlers (like Opentrons) that automate the "Make/Build" DNA assembly part of the cycle, reducing human error and feeding physical constructs straight into testing phases.

2. Computational & Software-Driven Design Hubs (In Silico DMTA)

• OpenProtein (PoET Platform): Provides a no-code front-end and API interface for biologists to upload data, run evolutionary transformers (like the Protein Evolutionary Transformer, PoET, and ESM-based models), and digitally build/analyze sequence libraries before physical synthesis.
• Ailurus Bio: Utilizes unsupervised pre-trained language models (like ESM-2) for intelligent library design coupled with automated downstream lab execution to optimize enzyme traits through closed feedback loops.

3. Enterprise & Commercial Closed-Loop Foundries

• Ginkgo Bioworks: Operates a massive horizontal automated foundry platform that programs cells. Ginkgo integrates proprietary software design with automated physical code execution (strain engineering, protein production, and testing) to run continuous DMTA cycles for industrial and therapeutic partners.
• Refining & Specialty AI-Bio Companies (e.g., Generate Biomedicines, Absci, and Profluent): These companies operate proprietary, end-to-end wet-lab-integrated computational engines. They generate massive proprietary training datasets by designing novel proteins in silico , manufacturing them via high-throughput automated screening (Make/Test), and piping the functional assay readouts back into deep learning models (Analyze/Learn).

Would you like to explore how to set up an open-source software stack for the computational (Design/Analyze) portion, or are you more interested in the hardware/robotics side of automated biofoundries?

If by “design–make–test–analyze” you mean a genuinely iterative protein-engineering loop—where computational design proposes variants, the platform actually gets them built/tested, and experimental results feed the next design round—the landscape is still fairly small.

Strongest examples

• cradle.bio — Probably the clearest commercial software platform for the loop. Its AI designs protein variants, learns from users’ experimental results, and generates optimized libraries for subsequent rounds. Cradle also partnered with ginkgobioworks.com so Cradle-generated sequences could be automatically tested in Ginkgo's foundry and the resulting data fed back into the model.
• arzeda.com — Runs an explicitly agentic design–build–test–learn system for industrial proteins/enzymes, combining AI/physics-based design with its own experimental data and automation.
• generatebiomedicines.com — An integrated generative-biology platform for therapeutic proteins. Generate describes its system as a continuous generate → build → measure → learn loop, with experimental results continuously training/refining the models.
• DPTX (Data Powered Therapeutics) — A newer, explicitly closed-loop platform combining computational protein design with automated expression, purification and characterization. Its stated loop is Design → Build → Test → Learn, with assay data feeding directly back into the next generation.
• ZerothBIO / ProPEL — Building an autonomous protein-engineering architecture connecting protein design, DNA-library construction, cell-free protein synthesis, functional measurement and learning. It is particularly focused on generating high-quality experimentally measured protein-function data.
• HemeAI — Focused on industrial enzymes. It combines protein-language models, automated expression/assays and active learning; experimental data feeds Bayesian optimization for the next synthesis round.
• RunLab — A newer computational protein-design platform that explicitly closes the loop through wet-lab validation and model learning, targeting enzymes and therapeutic binders.

Important distinction

There are really three levels of “closed loop”:

There are also academic/self-driving-lab implementations. For example, Northwestern's DREAM Cloud Lab is an NSF-supported protein-engineering facility built specifically around an AI-driven DBTL cycle, using automated cell-free protein synthesis and functional characterization. Center for Synthetic Biology A 2024 research platform similarly demonstrated four automated protein-evolution rounds in 10 days using a protein language model + biofoundry + machine-learning feedback loop.

If you're asking because you're mapping the competitive landscape, I'd put Cradle, Ginkgo, Generate, Arzeda, DPTX, and ZerothBIO in the first tier to investigate. The interesting question isn't merely who uses AI for protein design, but who owns/controls the experimental loop and automatically turns each assay into the next design decision. That's a much smaller set.

Running the Design-Make-Test-Analyze (DMTA) loop—often referred to as AI-driven lab-in-the-loop or self-driving laboratories for protein engineering—requires tight integration between generative machine learning models and high-throughput automated wet labs.

Several commercial platforms, specialized biofoundries, and open-source/academic architectures execute this continuous loop:

• Adaptyv Bio — Offers an automated robotics platform and foundry specifically built for protein design validation. Users submit sequence designs through their portal, and the automated lab executes cell-free protein expression and binding/affinity assays (returning data in days), functioning as a direct "Test" engine that feeds back into design loops. They also host Proteinbase , an open hub for experimental protein design and negative assay data.
• Cradle — Provides a software-driven protein engineering platform that uses AI (similar to large language models) to suggest sequence variants optimized for properties like expression, stability, and binding. Cradle links its generative modeling software with high-throughput wet-lab generation to iteratively retrain custom predictors on a project-by-project basis.
• Nabla Bio — Operates an integrated platform centered around JAM (Joint Atomic Modeling), a multimodal generative AI system. Nabla tightly couples its high-throughput mammalian expression and functional/binding assays with its modeling pipeline to generate human-relevant training data, bypassing iterative manual bottlenecks for antibody and de novo design.
• Academic & Open-Source Architectures (e.g., SAMPLE) — Frameworks like SAMPLE (Self-driving Autonomous Machines for Protein Landscape Exploration) represent the open-source blueprint for closed-loop systems. SAMPLE pairs active learning/Bayesian optimization software agents directly with robotic cloud lab infrastructure (such as Strateos) to autonomously cycle through sequence selection, expression, and thermal stability assays without human intervention.
• Enterprise Biofoundries (e.g., Ginkgo Bioworks & Ginkgo/AWS ecosystem) — Large-scale commercial biofoundries utilize massive robotic tracks to automate strain engineering, protein production, and screening. By integrating automated analytics pipelines with custom machine learning scoring engines, they run massive parallel DMTA cycles for industrial enzymes and biotherapeutics.

Would you like to explore how to connect your own software models to an external testing foundry, or are you looking for details on specific assay types (like binding affinity vs. thermal stability) supported by these platforms?

Several biotechnology companies and automated infrastructure providers run closed-loop Design-Make-Test-Analyze (DMTA) (or Design-Make-Test-Learn) workflows for protein engineering. These platforms merge generative artificial intelligence (AI) with high-throughput robotics and automated wet-lab assays to iterate rapidly on protein design.

Key Platforms and Companies

• Adaptyv Bio : Operates an automated protein engineering foundry and cloud-lab platform that lets researchers test AI-designed proteins. Their infrastructure automates binding affinity, expression, and thermostability assays, feeding real-world biophysical data straight back into digital design loops.
• Generate Biomedicines : Utilizes a machine-learning-powered generative biology platform (The Generative Biology Platform ) that creates novel clinical-stage protein therapeutics—such as antibodies, enzymes, and peptides—by tightly integrating computational design with high-throughput automated screening and validation.
• Amgen (Generative Loop Infrastructure) : Built an internal, highly scaled automated platform combining machine learning models with automated robotics to execute the full design-make-test-learn cycle, which reportedly cuts biologic discovery timelines in half.
• Profluent : Combines state-of-the-art protein language models (like OpenCRISPR) with functional validation pipelines to design, generate, and test genome-editing proteins and custom therapeutics through iterative computational-experimental feedback.
• Ginkgo Bioworks : Operates a massive horizontal automated foundry for synthetic biology that handles the "make and test" phases at industrial scale, frequently collaborating with AI protein designers to close the experimental loop.
• Prescient Design (Genentech/Roche) : Functions as an advanced AI-driven antibody and protein design platform integrated directly into Genentech’s therapeutic pipelines to transition computationally generated candidates straight into structural analysis and wet-lab screening.

Would you like to explore how automated cloud labs integrate with open-source protein design models , or focus on a specific therapeutic modality like antibodies versus enzymes?

The design–make–test–analyze (DMTA) loop in protein engineering is a closed-loop workflow where computational models propose protein variants (design), DNA/protein constructs are produced (make), experiments measure properties (test), and the results are used to guide the next design cycle (analyze/learn). It is closely related to the more common DBTL (Design–Build–Test–Learn) terminology.

Platforms that run some or all of this loop include:

A useful way to categorize them:

1. AI-first protein design platforms (mostly "Design")

These generate or rank candidate sequences:

• EvolutionaryScale
• Profluent
• Generate Biomedicines
• OpenProtein.AI
• ProGene

2. Autonomous wet-lab / biofoundry platforms (mostly "Make + Test")

These automate the physical loop:

• Adaptyv Bio
• DPTX
• other automated protein screening labs

3. Integrated closed-loop systems (closest to true DMTA)

These connect models, experiments, and feedback:

• DPTX — explicitly positions itself as a continuous design–build–test–learn system.
• Bigo.bio — integrates design, validation, and return of experimental results.
• ProteusAI — provides ML-guided DBTL tooling.
• Protein CREATE (academic platform) — combines AI design with high-throughput experimental feedback for binder engineering.

A caveat: a fully autonomous DMTA loop (AI designs → robots build → assays test → AI learns → repeat without much human intervention) is still emerging. Many commercial platforms currently cover only parts of the loop or integrate with external CROs/robotic labs rather than operating the entire stack internally.

Running the Design-Make-Test-Analyze (DMTA) —or Design-Build-Test-Learn (DBTL) —loop for protein engineering requires tightly coupling generative AI/machine learning with high-throughput laboratory automation. This "lab-in-the-loop" ecosystem bridges digital sequence/structure generation with physical synthesis, screening, and computational feedback.

The key technology platforms, biopharma infrastructures, and autonomous foundry systems driving these closed-loop cycles include:

1. Autonomous Foundries & Cloud Labs

• Ginkgo Bioworks: Utilizes massive robotic foundries combined with proprietary automation software (like Catalyst) and reconfigurable automation carts. Ginkgo integrates frontier AI reasoning models (such as collaborations with OpenAI) to autonomously orchestrate cell-free and cellular protein synthesis, running tens of thousands of experimental conditions in continuous closed loops.
• HighRes Biosolutions: Specializes in the physical orchestration layer that enables closed-loop DMTA. Their integrated robotic workcells and scheduling software automate the high-throughput handling of microplates, liquid dispensing, and assay instrumentation used by major pharma to execute physical steps without human bottlenecks.

2. Biopharma & Proprietary AI-Driven Pipelines

• Amgen: Has engineered an internal generative loop combining machine learning models for protein design with robotic high-throughput labs to synthesize and assay biologic candidates at scale, drastically shrinking optimization timelines.
• Recursion Pharmaceuticals / Valence Discovery: Integrates massive automated biology data generation with deep learning architectures, scaling iterative loops from small molecules into complex biological modalities.

3. Open-Source & Modular Software Frameworks

If you are looking to assemble your own computational side of the loop rather than a full hardware foundry:

• ProteinDJ: A modular, open-source framework designed to streamline and automate protein binder design workflows by chaining structural tools like RFdiffusion (for backbone generation) and ProteinMPNN (for sequence assignment).
• Bindcraft: Uses an iterative structure-prediction and hallucination approach to optimize de novo protein binders, functioning as an automated "Design/Analyze" engine prior to physical validation.

If you are looking to build or implement a specific loop, let me know:

• Are you focusing on computational design software (like RFdiffusion/ProteinMPNN) or physical laboratory automation?
• What specific protein property are you trying to optimize (e.g., thermostability, binding affinity, expression yield)?

If by design–make–test–analyze you mean a genuinely iterative protein-engineering loop—where experimental results feed back into the next round of sequence design—the landscape is fairly concentrated.

Strongest fits

• Cradle — Probably the clearest software platform example. You provide experimental data, Cradle builds project-specific models, generates variants/plates, and uses each experimental round to improve subsequent designs. It supports enzymes, peptides, and antibodies.
• Generate:Biomedicines — Runs a vertically integrated generate → build → measure → learn loop. Its wet lab and computational systems are integrated, so this is more of a biotech/platform-as-a-service model than software you install in your lab.
• Arzeda — Particularly strong for industrial enzymes. Its Intelligent Protein Design Technology explicitly runs an AI-driven design–build–test–learn cycle, incorporating experimental results into subsequent design iterations.
• Codexis — Its CodeEvolver platform combines computational design, ML, bioinformatics and experimental validation in repeated engineering cycles. It's particularly focused on enzyme optimization rather than general-purpose protein design.
• Pando Bioscience — A more vertically integrated approach: AI designs enzymes, an ultra-high-throughput wet lab builds/screens them, and the resulting genotype–function data trains the next generation. The company describes rounds producing 10⁴–10⁶ data points.

Platforms that cover the broader DBTL infrastructure

• TeselaGen — An enterprise platform connecting sequence design, experiment planning, laboratory execution, data capture and ML analysis. It's broader than protein engineering but explicitly covers the complete DBTL cycle.
• ProteusAI — Open-source software covering protein discovery, structure-based design, prediction and ML-guided directed evolution. It's closer to the computational/analysis side of the loop; it doesn't provide the same integrated robotic wet lab as Cradle or Generate.
• Profluent — Foundation-model-driven protein design coupled to an in-house wet lab, with experimental results informing its design systems. More of an AI protein-authoring/partnership platform than a general DBTL SaaS product.

A useful way to segment them

One distinction matters: “AI protein design” ≠ “closed-loop protein engineering.” Companies such as EvolutionaryScale/ESM, NVIDIA/GenSLM-type systems, etc. can generate or predict proteins, but unless experimental results are systematically returned to the design algorithm, they aren't really running the full DBTL loop.

If you're asking this for competitive landscape / company diligence, I can also give you a more exhaustive 20–30 company map of protein-engineering DBTL platforms, classified by AI design, automated wet lab, assay type, closed-loop status, therapeutic vs enzyme focus, and whether it's SaaS vs outsourced R&D.

The Design-Make-Test-Analyze (DMTA) loop—often referred to as a "lab-in-the-loop" or self-driving lab workflow—is increasingly automated by combining generative artificial intelligence (AI) with robotic wet-lab infrastructure.

Platforms executing this closed-loop protein engineering span software orchestration frameworks, integrated biopharma internal engines, and robotic biofoundries:

• Integromat/Orchestration & Lab-in-the-Loop Software:
  • NVIDIA BioNeMo / NIM: Used as a foundational AI layer integrated with physical laboratory systems and digital twins to orchestrate real-time data capture, simulation, and closed-loop feedback for molecular and protein discovery.
  • Open-source/Modular Frameworks (e.g., ProteinDJ): Modular frameworks designed to streamline steps like de novo binder generation (combining tools like RFdiffusion and ProteinMPNN ) before handoff to physical synthesis.
• Enterprise & Biopharma Integrated Platforms:
  • Amgen (AMPLIFY & Internal Generative Platforms): Uses custom co-developed protein language models (like AMPLIFY with Mila) tightly integrated with robotic automation to execute a continuous design-make-test-learn loop, drastically cutting discovery timelines.
  • Other major biopharma and synthetic biology companies (such as Ginkgo Bioworks and Recursion/Exscientia ecosystems) deploy massive high-throughput automated foundries that feed physical expression and characterization data directly back into machine learning design models.
• Academic & Infrastructure Biofoundries:
  • Global biofoundries (part of the Global Biofoundries Alliance) utilize integrated liquid handlers, automated microplate readers, and analytical instruments (like high-throughput mass spec and chromatography) tied to LIMS (Laboratory Information Management Systems) software to cycle through DMTA iterations autonomously.

If you are looking for platforms for a specific project, let me know:

• Are you looking for open-source software to run locally, or cloud/commercial enterprise platforms?
• Are you focused on a specific application like de novo binder design, enzyme optimization , or antibody engineering?

Yes. If by “design–make–test–analyze” you mean a genuinely iterative loop where experimental results automatically influence the next protein designs—not merely a protein-design model—there are several notable platforms.

The strongest examples

• Cradle — Probably one of the clearest commercial examples. Cradle generates protein variants, supports experimental testing, automatically learns from project-specific assay data, and uses that information for subsequent rounds. It also has a wet-lab operation.
• Generate:Biomedicines — Its Generate Platform explicitly implements a continuous Generate → Build → Measure → Learn cycle. It combines generative protein design with scaled biological production and measurement, with experimental data feeding the next generation of designs.
• Ginkgo Bioworks / Cradle — Ginkgo's Foundry provides the automated build/test side, while Cradle provides AI protein design. Their integration explicitly sends Cradle-designed sequences into Ginkgo for automated testing and returns the experimental data to Cradle's models.
• Arzeda — Runs a particularly interesting AI-agent-driven closed-loop DBTL system. Its platform combines physics-based design, proprietary experimental data, automated testing, and feedback into subsequent design cycles.
• Imperagen — Focused specifically on enzyme optimization. It describes an integrated loop in which physics-based models select variants, hundreds are automatically built/tested, and the results directly update the AI model for the next round.
• RunLab Bio — An AI-first protein-design factory with generative design, multi-objective ranking, wet-lab validation, and model learning from every cycle.
• ZerothBIO — Its ProPEL system explicitly connects protein design, cell-free production, functional testing, and learning into a closed loop.
• Data Powered Therapeutics (DPTX) — Describes an end-to-end automated Design → Build → Test → Learn system encompassing computational design, expression, purification, biophysical testing, and feedback into models.

Important distinction

There are really three tiers here:

The last distinction matters: ProteusAI, for example, is an open-source software platform covering several stages of DBTL, but it isn't itself an autonomous wet-lab factory.

There is also an emerging academic ecosystem of self-driving protein-engineering labs. A 2026 review describes these as systems coupling automated library design, robotic transformation/expression, assays, and ML-driven decision-making into closed optimization loops.

If you're mapping the competitive landscape, I'd put Cradle, Generate, Arzeda, Imperagen, Ginkgo/Cradle, and RunLab in the “commercial closed-loop protein engineering” bucket, with ProteusAI and academic biofoundry systems in the “software/research infrastructure” bucket.