Scope and Limitations

This handbook is for biological research, biotechnology, and technical evaluation. It is not medical advice, public health guidance, legal advice, biosafety guidance, or regulatory advice. The scope is AI across molecular discovery and design, cells, tissues, systems biology, organismal and environmental biology, therapeutic translation, and the research systems that connect them. Adjacent domains (clinical AI, public health AI, biosecurity) are covered by companion handbooks in the same series.

In Scope

The handbook covers AI methods, evaluation frameworks, and institutional infrastructure for biological discovery and biotechnology:

• AI models for biological data. Protein sequence and structure, biomolecular interactions, single-cell and spatial transcriptomics, image-based phenotyping, genome and regulatory sequence, plant and crop data, ecological observations, neural recordings, aging biomarkers, small-molecule chemistry, antibody and biologic design, perturbation prediction.
• Cells, tissues, and systems biology. Cell-state models, tissue atlases, image phenotyping, perturbation biology, systems biology, gene regulatory networks, and multiscale modeling.
• Organismal and environmental biology. Neuroscience AI, aging and longevity biology, plant and crop AI, environmental DNA, biodiversity monitoring, conservation biology, and ecological modeling.
• Therapeutic discovery and translation. Target identification, protein and antibody design, small-molecule generation and ADMET prediction, chemical biology, vaccine and RNA design, cell and gene therapy, diagnostics and biomarkers, clinical-trial AI for translational research, translational evidence and failure modes.
• Research systems and automation. Self-driving laboratories, cloud-lab infrastructure, synthetic biology design tools, biomanufacturing, agentic science workflows, with evidence anchored to bounded demonstrations such as Coscientist (Boiko et al., 2023) and Virtual Lab (Swanson et al., 2025).
• Practice and governance. Benchmarks, reproducibility and open science, information hazards in capability research, workforce and compute infrastructure, institutional readiness.
• Evaluation infrastructure. Five-question chapter utility standard, biology-aware splits, calibration, validity checks, and the credibility hierarchy from blinded benchmarks down to self-reported numbers. DOME reporting (Walsh et al., 2021) and PoseBusters-style validity checks (Buttenschoen et al., 2024) are representative standards.

Out of Scope

The handbook does not cover the following topics, even when they touch life sciences AI:

• Patient-specific medical advice. Decisions about individual patient care belong with treating clinicians and the clinical AI resources designed for that purpose.
• Operational biosafety protocols. Detailed biosafety procedures (BSL-1 through BSL-4 operation, select agent handling, decontamination procedures) belong in biosafety manuals and institutional biosafety committee documentation.
• Step-by-step instructions for biological misuse. The Information Hazards in Capability Research chapter discusses the framework for handling such information; the handbook itself does not include operational misuse content.
• Product endorsement. Named systems are described as canonical published examples of capability classes, not as recommendations.
• Legal or regulatory determinations. The handbook references FDA AI/ML drug-development materials (FDA, 2026), EMA AI lifecycle materials (EMA, 2024), and ARPA-H program pages (ARPA-H IGoR, 2026) but does not provide regulatory or legal advice.
• Farm operations and environmental-policy implementation. Plant and ecology chapters cover biological modeling and evidence quality, not farm management, commodity markets, pesticide-use advice, land-use law, or environmental permitting.
• Every model release. The handbook focuses on major and demonstrated systems rather than attempting to enumerate every release.

Adjacent Handbooks for Related Scope

The Life Sciences AI Handbook is the discovery layer in a series with three companion handbooks that cover adjacent layers:

• Clinical AI. For AI in patient care, FDA-cleared diagnostic tools, clinical decision support, AI-assisted documentation, medical liability, and clinical workflow integration.
• Public health AI. For AI in disease surveillance, epidemic forecasting, genomic pathogen analysis, health-department implementation, and population-level evaluation.
• Biosecurity. For biological risk, the Biological Weapons Convention, dual-use research oversight at depth, DNA synthesis screening operations, AI-enabled pathogen design risk, and governance frameworks for AI-bio convergence.

Cross-references between handbooks appear on the welcome page series cross-links. Each handbook is self-contained for its scope.

Claim Calibration

After the field-guide layer, the handbook uses five recurring questions to prevent unsupported certainty:

• What is demonstrated? Claims supported by peer-reviewed evidence, official documentation, or reproducible blind benchmark results. The evidence must be specific to a defined task and dataset.
• What is theoretical? Claims plausible given current methods but not yet established for routine use.
• What is beyond current capability? Claims not supported by credible evidence with current systems.
• What would make this more promising? The next evidence that would make the area stronger, expose limits, or narrow the promise.
• What should researchers, biotech teams, funders, and program leaders do with this? The practical decision implied by the evidence.

When evidence is mixed, the text favours the lower-confidence tier. Self-reported numbers from companies and labs are weighted below independent reproductions. Vendor metrics are labeled “reported” or “company-reported” unless independently validated. Preprints are labeled and cited as preprints. This discipline matters because independent follow-up can change the interpretation of an apparently successful result, as in the A-Lab novelty debate (Szymanski et al., 2023; Leeman et al., 2024). The intent is not to be conservative; the intent is to be specific.

Sources and Verification

Citations are checked against Crossref, PubMed, publisher records, or official organizational sources where available. Where a system has a peer-reviewed publication and a separate preprint, the peer-reviewed version is preferred and the preprint history is noted. Where a system is restricted-release, such as AlphaFold 3 initial release (Abramson et al., 2024) or AlphaProteo (Zambaldi et al., 2024, preprint), the restriction is named explicitly.

When citation errors are found (a chimera, a wrong-author attribution, a journal mix-up), they are corrected and noted. Citation discipline is part of the evidence standard.

Update Cadence

The handbook is continuously updated as major papers and benchmark results land in Nature, Science, Cell, and adjacent venues, and as frontier-lab and regulatory actions warrant. Each chapter carries the date of last substantive update. The handbook is maintained as a working reference, not as a frozen edition.

License

The handbook is open access under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Citation, adaptation, and reuse are encouraged with attribution. See the How to Cite page for citation formats.

Disclaimer

The handbook is for educational, research, and technical-evaluation purposes. It is not medical advice, public health guidance, biosafety operations guidance, regulatory advice, or legal counsel. Model outputs discussed here are hypotheses that require experimental validation; they are not endpoints.

The handbook supports careful reading. It does not replace professional judgement, institutional review, regulatory engagement, or domain-specific expertise.