AI for the Life Sciences

Published

August 22, 2026

Life sciences AI is the application of machine learning systems to biological objects: sequences, structures, cells, tissues, organisms, experiments. Its center of gravity is upstream of clinical care: molecules generated before they are synthesized, cells represented before they are perturbed, experiments planned before they are run. AlphaFold 2 made over 214 million predicted structures freely available to researchers (Varadi et al., 2024). RFdiffusion produced experimentally tested protein designs (Watson et al., 2023). Coscientist used an LLM with chemistry and hardware tools to plan and execute bounded chemical procedures (Boiko et al., 2023). None of this removes the experiment. The work is choosing the right one.

Learning Objectives

Use this chapter to:

  • Define life sciences AI as the discovery layer for molecules, cells, tissues, organisms, environments, experiments, and research systems.
  • Explain why model output remains a hypothesis until a benchmark, experiment, or deployment context gives it evidentiary weight.

Prerequisites: none for orientation. The remaining chapters assume familiarity with this chapter and the Executive Summary.

Summary: Define life sciences AI as the discovery layer for molecules, cells, tissues, organisms, environments, experiments, and research systems. The field has mature anchors in structure prediction and protein design, faster-moving evidence in cells and genomes, and earlier claims in autonomous discovery.

Key point: Model output is treated as a hypothesis until a benchmark, experiment, or deployment context gives it weight. Open question: how often those systems improve experiments outside the tasks where they were built.

Bottom line: The chapter sets the grammar for the rest of the handbook: biological object, task, evidence, confidence, and downstream decision.

Field Guide

What is this field trying to solve? Define life sciences AI as the discovery layer for molecules, cells, tissues, organisms, environments, experiments, and research systems.

What is the core idea? Model output is treated as a hypothesis until a benchmark, experiment, or deployment context gives it weight.

What is the current state of the field? The field has mature anchors in structure prediction and protein design, faster-moving evidence in cells and genomes, and earlier claims in autonomous discovery.

What do we know, and what remains open? Known reference points include AlphaFold, RFdiffusion, Evo, scGPT, Geneformer, GEARS, Coscientist, Virtual Lab, CASP, PoseBusters, DOME, and biology-aware benchmark suites. What remains open is how often those systems improve experiments outside the tasks where they were built.

Why does this matter? The chapter sets the grammar for the rest of the handbook: biological object, task, evidence, confidence, and downstream decision.

Introduction: What This Handbook Is For

October 2024, Stockholm: The Royal Swedish Academy of Sciences awarded one half of the Nobel Prize in Chemistry to David Baker for computational protein design and the other half jointly to Demis Hassabis and John Jumper for protein structure prediction (Nobel Prize, 2024). The award recognized the scientific impact of computational protein design and AlphaFold.

May to November 2024: AlphaFold 3 extended biomolecular prediction beyond proteins to nucleic acids, ions, and ligands (Abramson et al., 2024). The initial server-only release drew reproducibility criticism. DeepMind released inference code and weights for academic, noncommercial use in November 2024, documented in a Nature addendum (Abramson et al., 2024). Boltz and Chai were separately developed under different licences.

September 2024 to October 2025: OpenAI reported o1 performance on GPQA Diamond, a graduate-level multiple-choice benchmark spanning biology, chemistry, and physics (OpenAI, September 2024). This vendor-reported benchmark does not measure wet-lab performance or biological discovery. Anthropic later announced Claude for Life Sciences with integrations into Benchling, 10x Genomics, PubMed, and Synapse (Anthropic, October 2025).

May 2026: ARPA-H launches the IGoR program (Intelligent Generator of Research) to “deliver gold-standard biomedical science faster through an AI-powered research ecosystem” focused on Alzheimer’s, Parkinson’s, and autoimmune disease (HHS press release, May 2026).

June 2026: Anthropic launches Claude Science as a beta AI workbench for scientific research, describing Claude as an interface for literature search, data analysis, code execution, remote compute, and scientific software integrations (Anthropic, June 2026).

Together, these events changed the workflow of biological research: AI is no longer only an experiment to run, it is infrastructure to evaluate and use. The question is no longer “can I use AI?” but “which model, for which decision, with what validation?”

This handbook is written for those questions. It is not a tour of every model release. It is a framework for reading model outputs as research inputs, choosing systems by biological question rather than brand, and designing the experiment that turns a prediction into biological evidence.

What Life Sciences AI Is, and What It Is Not

The Discovery Layer

Life sciences AI sits in the discovery layer of biological research:

                     Population health
                            │
                            ▼
  Public Health AI ── Patient populations, surveillance, forecasting
                            │
                            ▼
                     Clinical practice
                            │
                            ▼
       Clinical AI ── Diagnosis, treatment, workflow, liability
                            │
                            ▼
                   Therapeutic development
                            │
                            ▼
   [LIFE SCIENCES AI] ── Molecules, cells, organisms, environments,
                          experiments, candidates
                            │
                            ▼
                       Fundamental biology

Each layer above depends on inputs from the layer below. A drug candidate exists before a clinical trial; a clinical trial exists before a regulatory decision; a regulatory decision exists before a population-scale deployment. Life sciences AI operates across molecules, cells, tissues, organisms, environments, and experiments. Its outputs can feed clinical, public health, biotechnology, agricultural, ecological, and biosecurity decisions.

The Three Adjacent Domains

Domain Object of Study Failure Cost Evidence Standard
Life sciences AI Molecules, cells, tissues, organisms, environments, experiments, research decisions A failed experiment, a discontinued program, a non-validated paper Benchmarks + prospective experimental validation
Clinical AI Individual patients, diagnostic and therapeutic decisions Misdiagnosis, mistreatment, liability Prospective clinical trials, FDA clearance, real-world performance
Public health AI Populations, surveillance, intervention design Missed outbreaks, misallocated resources, eroded trust Deployment-context evaluation, proper scoring rules, equity analysis

The three are complementary but distinct. A model that predicts a binding interaction is a life sciences AI claim. A model that recommends a treatment for a specific patient is a clinical AI claim. A model that forecasts hospitalizations is a public health AI claim. The same architectural family (transformers, diffusion, graph networks) underlies all three, but the evidence standard depends on the decision the model informs, not the math under the hood.

Companion handbooks in this series cover the clinical, population, and biosecurity layers explicitly. Links and descriptions appear on the welcome page under “Companion Handbooks.”

The Chapter Reference and Utility Standard

Every major chapter in this handbook starts with a field guide, then calibrates evidence. The reference layer answers:

  • What is this field trying to solve?
  • What is the core idea?
  • What is the current state of the field?
  • What do we know, and what remains open?
  • Why does this matter?

The evidence layer answers:

What is demonstrated?

Supported by published evidence in peer-reviewed venues, official documentation, or reproducible benchmark results. The evidence must be specific to a defined task and dataset. “AlphaFold 2 predicts single-chain protein structures at near-experimental accuracy for the majority of well-folded domains evaluated in CASP14” is a demonstrated claim (Jumper et al., 2021).

What is theoretical?

Plausible given current methods but not yet established for routine use. The capability has been shown in selected systems, narrow tasks, or controlled settings without proven generalization. “Single-cell foundation models can transfer to new tissues and species” is currently a theoretical claim: published evidence supports transfer in some settings (Cui et al., 2024; Theodoris et al., 2023), but the boundary of useful transfer is an open research question.

What is beyond current capability?

Not supported by credible evidence with current systems. The capability is either aspirational, has been demonstrated in toy settings that do not generalize, or requires evidence that has not been produced. “Fully autonomous drug discovery without experimental validation” is beyond current capabilities. Coscientist demonstrates autonomous chemistry execution in bounded settings (Boiko et al., 2023), not autonomous discovery without measurement.

What would make this more promising?

The answer should name the next measurement: a blinded benchmark, prospective wet-lab validation, independent reproduction, field study, external cohort, or decision-relevant endpoint. A claim that does not name the missing evidence is not ready for program decisions.

What should researchers, biotech teams, funders, and program leaders do with this?

The answer should translate evidence into action: evaluate, adopt narrowly, test experimentally, fund as infrastructure, fund as research, or defer. The handbook treats action as part of evidence reading because science programs spend money, time, samples, and credibility on these decisions.

The point of the questions is not to be conservative: it is to be specific. A claim that cannot answer them is not a claim about a system; it is marketing.

Reading Claims with the Framework

When a press release, a paper title, or a vendor pitch makes a capability claim, ask:

  1. What biological object is the model representing? Sequence? Structure? Cell state? Tissue? Experiment? Reaction?
  2. What is the specific task on which the claim is made? Property prediction? Generation? Ranking? Classification? Planning?
  3. What is the evidence? Held-out benchmark performance? Prospective experiment? Cross-laboratory replication? Vendor-reported internal evaluation?
  4. What would make this more promising? What evidence is still missing?
  5. What should researchers, biotech teams, funders, and program leaders do with this? Evaluate, adopt narrowly, test, fund, or defer?

A claim that survives these questions is at minimum demonstrated for the specific task and dataset. A claim that cannot name the missing evidence is either theoretical or beyond current capability.

The Capability Landscape

The handbook organizes life sciences AI into seven parts, followed by a closing chapter on emerging frontiers. Each part applies the five-question standard to a defined layer of research or practice.

Part I: Foundations (this part)

Part II: Molecular Discovery and Design

Part III: Cells, Tissues, and Systems Biology

Part IV: Organismal and Environmental Biology

Part V: Therapeutic Discovery and Translation

Part VI: Research Systems and Automation

Part VII: Evaluation, Practice, and Governance

Closing Perspective

The Infrastructure Layer

Models are visible; infrastructure is decisive. The capability gaps in life sciences AI are often data gaps, benchmark gaps, or compute gaps before they are architecture gaps.

Public Programs

  • NIH Bridge2AI (NIH Common Fund, Bridge2AI Consortium): Four grand-challenge data generation projects (CHORUS for AI/ML in clinical care, CM4AI for functional genomics, VOICE for precision public health, AI-READI for salutogenesis). The program’s premise: AI-ready datasets are themselves an infrastructure problem, requiring metadata, ethics review, quality control, and workforce development: not only more storage.
  • ARPA-H IGoR (ARPA-H programs page; HHS press release, May 2026): Intelligent Generator of Research, focused on Alzheimer’s, Parkinson’s, and autoimmune disease. ARPA-H also funds adjacent AI programs: ADVOCATE (cardiovascular AI agents), RAPID (rare-disease AI diagnostics), CATALYST (ADME-tox modeling), ADAPT (precision cancer therapy).
  • NCI Cancer Research Data Commons (NCI CRDC): Data infrastructure spanning genomics, proteomics, and imaging that AI work depends on, even when not formally an “AI program.”

Non-Profit and Foundation Programs

  • CZ Biohub and CZ CELLxGENE (CZ CELLxGENE Discover): Roughly 100 million curated single-cell observations in a standardized, queryable platform. The Tabula Sapiens collection (1.1M cells from 28 organs, 24 donors) is a benchmark first-draft human cell atlas.
  • CZI Virtual Cells Platform (CZI): An active program to build and benchmark foundation models for cell biology.
  • Arc Institute: Co-developer (with Stanford, UC Berkeley, UCSF, and NVIDIA) of the Evo and Evo 2 genomic foundation models (Nguyen et al., 2024; Brixi et al., 2026).

Frontier Labs

The major AI labs each have life-sciences programs at varying degrees of openness:

Lab Visible Life-Sciences Work Evidence Type
Google DeepMind / Isomorphic Labs AlphaFold 2/3, AlphaMissense, AlphaProteo; Eli Lilly and Novartis drug-discovery partnerships (Isomorphic Labs, January 2024) Peer-reviewed for AlphaFold lineage; AlphaProteo is arXiv preprint (Zambaldi et al., 2024, preprint); partnerships are factual but not efficacy evidence
Anthropic Claude for Life Sciences (October 2025), Claude Science beta (June 2026), AI for Science Program Company announcements and official product materials; no peer-reviewed life-sciences outcome study identified by the August 12, 2026 review cutoff
OpenAI Color Health cancer-screening copilot (OpenAI, June 2024); Moderna ChatGPT Enterprise deployment (OpenAI, April 2024); o1 model GPQA Diamond performance (OpenAI, September 2024) Company-reported partnerships and benchmark results; not peer-reviewed evidence of biological discovery
Meta AI and EvolutionaryScale Meta AI developed ESM-2 (Lin et al., 2023); EvolutionaryScale developed ESM3 (Hayes et al., 2025) Both peer-reviewed; distinct organizations and model releases

Drug Discovery Companies Using AI

  • Recursion Pharmaceuticals (Recursion pipeline): High-content imaging and machine learning used within a therapeutic-development platform. Candidate status and company-reported clinical signals should be verified against the current pipeline and the corresponding trial record rather than treated as platform-level efficacy evidence.
  • Rentosertib (ISM001-055): The discovery program and Phase I work were reported in Nature Biotechnology, followed by a randomized Phase 2a trial in Nature Medicine (Ren et al., 2025; Xu et al., 2025). The primary Phase 2a endpoint was safety and tolerability; efficacy measures were exploratory.
  • Insitro: Machine-learning models for metabolic disease and neuroscience; expanded Eli Lilly small-molecule collaboration in September 2025.

Read these with the framework: a partnership announcement is factual evidence of the partnership; it is not evidence that the AI-discovered molecule will read out positively, advance to Phase III, or change a patient’s outcome.

Who This Handbook Is For

The handbook is written for several overlapping audiences. The shared question is: when does an AI output deserve experimental attention?

Role Primary Need From This Handbook
Computational biologist Capability tier for each model class; failure modes; benchmark design matched to the scientific question
Biotechnology team lead Build-vs-buy framing; license diligence; what the open-source alternatives are when a frontier release is restricted
Drug discovery scientist Where AI shifts a stage gate vs. where it does not; how to read a vendor pitch against published evidence
Physician-scientist Translation between bench AI and clinical decision-making; what makes a discovery-stage AI claim relevant to the clinic
Synthetic biologist Design tools, autonomous lab integration, dual-use considerations
Graduate student Conceptual entry points into model classes; canonical citations; how to read benchmark results
Research program leader Capital allocation framing; which capabilities are infrastructure-grade vs. research-grade; how to evaluate proposals that invoke AI

How to Read the Rest of the Handbook

Quick Orientation

  1. Executive Summary: Handbook-wide conclusions
  2. Protein Structure Prediction: The best-established capability and its limits
  3. Evaluation Principles for Life Sciences AI: The framework that turns capability into decision

Deeper Orientation

Add:

  1. Single-Cell Foundation Models: The capability frontier in cell biology
  2. Self-Driving Laboratories: The autonomous laboratory frontier
  3. Information Hazards in Capability Research: Dual-use considerations for design and generation tools

Deep program review

Read the relevant Part end-to-end. Each chapter is self-contained but cross-references the others.

Common Questions

Is life sciences AI the same as “AI for science”?

Overlapping but not identical. “AI for science” is a broader phrase that includes physics, chemistry, materials, climate, mathematics, and other domains. This handbook focuses on biological systems: proteins, cells, genomes, tissues, organisms, environments, and the research workflows that connect them.

Do I need to know machine learning to use this handbook?

No. The handbook is written for working biologists, drug discovery scientists, and program leaders. It treats ML as a research instrument, analogous to a sequencer or microscope: the output and its limits must be understood, but readers need not build the instrument from scratch. The Foundation Models for Biology and Evaluation Principles chapters provide the ML context needed for the remaining chapters.

How often does the handbook update?

Continuously, as major papers and benchmark results appear in Nature, Science, Cell, and adjacent venues, and as frontier-lab and FDA actions warrant. The publication and modification dates are visible in each chapter’s metadata.

Why is dual-use treated as a separate chapter rather than woven through?

Because the questions in dual-use review (what to publish, what to release, how to communicate capability without enabling misuse) apply to many model classes at once. The dedicated chapter (Information Hazards in Capability Research) keeps the framework in one place. Cross-references in individual chapters point back to it.

What’s the difference between a foundation model and a task-specific model in biology?

A foundation model is pretrained on broad data (sequences, structures, cells) and adapted to many downstream tasks. A task-specific model is trained for one task. In life sciences, the foundation-model wave is real: ESM-2 / ESMFold for proteins, scGPT and Geneformer for single cells, Evo and Evo 2 for genomes. The wave is also recent and the generalization claims are still being validated. Foundation Models for Biology treats this in depth.

Are AI-discovered drugs real yet?

Some AI-associated molecules have entered clinical trials. Rentosertib is a peer-reviewed example, but its Phase 2a primary endpoint was safety and tolerability, not registration-level efficacy. Public evidence reviewed on August 12, 2026 did not identify an FDA- or EMA-approved small molecule whose molecular discovery was attributed to generative AI. Translational Evidence and Failure Modes covers the evidentiary boundary.

Demonstrated capability

Demonstrated life sciences AI is task-specific. AlphaFold 2 supports single-chain protein structure prediction for many well-folded domains; RFdiffusion supports experimentally tested protein design; AlphaMissense supports missense-variant triage as a research tool; scGPT, Geneformer, Evo, Evo 2, GEARS, Coscientist, and Virtual Lab support selected representation, prediction, generation, and laboratory-workflow tasks. Each result is demonstrated only within its evaluated biology, data distribution, and validation design.

Theoretical capability

Theoretical claims include routine transfer across unseen tissues, species, protein families, cell states, and experimental platforms; cross-modality systems that reason cleanly from sequence to cell to organism; and AI agents that plan biology programs with limited human supervision. Current methods make those claims plausible in bounded settings, but the evidence still depends on prospective validation, independent replication, and the specific decision being made.

Beyond current capability

Beyond current capability includes general biological reasoning that removes the need for experiments, fully autonomous drug or biology discovery without human accountability, and model outputs treated as biological facts. A prediction may be useful, but measurement remains the ground truth. Any claim that skips the falsifying experiment is outside the evidence standard used in this handbook.

Evidence that would change the assessment

The area becomes more promising with repeated prospective studies showing that AI-selected molecules, perturbations, targets, or experiments improve decisions across independent laboratories and biological contexts. Stronger evidence would pair benchmark performance with experimental denominators: what was generated, selected, tested, failed, and advanced. The decisive evidence is not a larger model announcement; it is a model-linked experiment that changes a research or program decision and survives external reproduction.

Implications for research and program decisions

  • Name the biological object first. Sequence, structure, ligand, cell state, tissue, experiment, or clinical endpoint. The right model class depends on the object.
  • Name the validation object second. Specify the experiment, benchmark, or independent dataset that would change the decision if the model were wrong.
  • Do not equate a model score with biological truth. A high pLDDT, a low Tanimoto, a strong attention weight: these are model outputs, not measurements.
  • Treat every vendor claim as a claim about a specific data distribution until proven otherwise. A model that works on one cell line, one species, or one assay does not work on all of them.
  • Read the license before scoping the project. A model that cannot run on the required infrastructure is not usable for the program.
  • Cite by version and venue. “AlphaFold” without a version is ambiguous; an announcement is not a paper; a partnership is not an outcome.

Cross-References