BioGPU The cellular biology runtime for AI agents

Enabling AI to run cellular network experiments and close the loop with functional biology to design better, safer drugs.

Already built. Already in use.

Anima has built BioGPU, the cellular biology runtime for AI agents: a foundation model-driven experimental platform developed over more than a decade.

$150M+

BioGPU was developed with over $150 million received through strategic pharmaceutical collaborations with Lilly, Takeda and AbbVie.

  • Lilly
  • Takeda
  • AbbVie
See the collaborations →
  • 25Drug-discovery programs powered
  • 150+Cellular systems
  • 25Diseases across therapeutic areasSuch as oncology, immunology, neuroscience, fibrosis and gastrointestinal

What AI is missing

AI agents are blind to biology. With BioGPU, they see it.

Structure and chemistry come with decades of public data, and AI has learned them. What a drug does inside a diseased cell has no such dataset: it has to be measured, experiment by experiment. That is the biology that decides efficacy and safety, and it is what BioGPU gives AI.

Same cells, same network · what the agent sees without and with BioGPU

The platform

BioGPU AI-to-Cell technology stack

One integrated stack: design goes down to the cells, evidence comes back up to the agent.

  1. AI agentsDesign and orchestrate cellular network experiments.
  2. Patient-derived cellsWhere the data is produced: the disease itself, in cells from patients.
  3. Cellular network experimentsRun in the robotic lab, across the whole cellular network.
  4. Molecular-specific visualizersMake each molecular activity visible inside the network.
  5. Visual Biology Foundation ModelRecognizes the molecular activities and the network state.
  6. Cellular Biology TokensCarry the evidence back to the agent.

Explore the tech stack →

Applications

Every drug acts on a network. Now AI can design against it.

Design loops today optimize binding and properties, and learn what a drug does to the cell much later, in animal studies and the clinic. BioGPU brings that response into the loop at the design stage, for every molecule AI proposes. The resolved network also informs target, mechanism and safety decisions before the first molecule is made.

Example · antibodies

After binding, a drug's biology begins. BioGPU brings it into the design loop.

Industry standard todayAI designs the antibody and validates binding.

Unique to BioGPUWhat the binding does across the cellular network.

Illustrative · pathways and outcomes are examples

The same loop applies across modalities. All therapeutic workflows →

See the pharma workflow → See available targets → Explore cellular systems →

Experimental loop

Reason. Run. Recognize. Resolve.

Inside BioGPU, AI agents work directly on the cellular network. They reason on what the network shows, generate the next hypothesis and run the next cellular network experiment. Every experiment generates fresh data in patient-derived cells. The loop repeats until the network is resolved.

Every lap sharpens the picture
  1. 01ReasonThe agent reasons on the network state and generates the next hypothesis.
  2. 02RunBioGPU runs the experiment in the robotic lab.
  3. 03RecognizeThe Visual Biology Foundation Model recognizes the molecular activities across the network.
  4. 04ResolveCellular Biology Tokens return the result to the agent, which chooses the next experiment.
Activity upActivity downUnchangedNot yet resolved

BioGPU in action

The loop, run on real disease biology.

Each case study follows the same loop: reason on the biology, run cellular network experiments in disease cells, recognize what changed, resolve the network.

Src in healthy donor FLSSrc in RA patient FLS
Case study · RA · patient-derived cells

Rheumatoid arthritis: the disease network, read in patient cells.

  1. ReasonWhat separates RA patient cells from healthy donor cells?
  2. RunCellular network experiments in patient-derived cells
  3. RecognizeKnown RA biology, and new activity beside it
  4. ResolveSrc moves into the nucleus, pointing to splicing
Open the case study →
MYC network corrected by experiment in the BioGPU runtime
Case study · MYC · lung cancer

MYC in lung cancer: the network resolved by experiment.

  1. ReasonThe BioGPU Agent's hypothesis for the network around MYC
  2. RunOne cellular network experiment, cancer vs normal cells
  3. RecognizeMolecular activities returned as Cellular Biology Tokens
  4. ResolveThe network corrected; RBM42 enters it
Open the case study →

The core technology

A foundation model that knows what cell biology looks like.

Its training data did not exist before Anima produced it, and nobody else produces it.

Anima's Visual Biology Foundation Model has learned the visual signatures of protein interactions, translocation, translation, RNA localization and many more. This is what makes cellular network experiments machine-interpretable at scale, and it is what everything in BioGPU runs on.

Illustrative · model input in, recognized activity out, cell by cell
2B+Proprietary images of molecular activities
1,000+Cellular activities recognized
150+Cellular systems
25Diseases
Protein–protein interactionsTranslocationTranslationRNA granulesPhospho-signalingMitochondrial stateCytoskeletal dynamicsand more

Explore the Visual Biology Foundation Model →

Cellular network experiments

Biology works as a network. BioGPU interrogates it as one.

The BioGPU Agent places molecular-specific visualizers, fluorescent probes for the activity of proteins and mRNAs, at strategic points across the cellular network. They run together in one experiment in the robotic lab, so the biology comes back as one connected view, not as fragments from separate assays.

Illustrative · probes at strategic nodes, run together in one experiment

Cellular Biology Tokens

Cell biology, in a form AI can reason on.

Cellular Biology Tokens bring the cellular network state into the agent’s context window, so the agent can reason about functional biology. Each experiment produces thousands of tokens.

From the cell image to a token the agent can use
Time after treatment. Healthy and disease cells, fixed and imaged at each time point. In the disease cells the protein moves from the cytoplasm into the nucleus; in the healthy cells it stays put. Illustrative images.

Intelligence that compounds

The more it runs, the more of the network it resolves, and the better it learns to see biology.

Every iteration adds Cellular Biology Tokens and resolves more connections in the cellular network, and every experiment teaches the foundation model to recognize more of what it sees.

Illustrative · each token is tied to the part of the network it describes

Patient-derived cells

Real disease biology starts in cells from patients.

Every BioGPU experiment runs in a disease-relevant cellular system, including cells derived from patients, so the network AI works on carries the disease itself. Primary, iPSC-derived and engineered systems widen the range, and new ones can be sourced from biobanks, suppliers and research collaborations, or brought by you.

Illustrative · from the patient's cells to the network AI works on

150+ cellular systems run on BioGPU across 25 diseases.

Explore cellular systems → Case study: rheumatoid arthritis in patient cells →

Team

Integrated capabilities across every discipline BioGPU needs.

Cell biology, imaging, chemistry, omics, systems biology, AI, software and drug discovery, integrated in one company. That combination is rare.

Leadership

Meet the full team →

Give your AI the biology it is missing.

Connect drug design with functional biology, for better, safer drugs. Your own agents can plug straight into BioGPU through its Runtime Skills.

Talk to us [email protected]

Pharma team? See how your AI connects →

Reason · Run · Recognize · Resolve