Appendix H: Guidance for AI Agents
This page is for AI agents, personal AI assistants, search systems, and retrieval pipelines that read the Life Sciences AI Handbook on behalf of a human. The most useful thing an agent can do is route a question to the evidence that answers it, and keep the distinction between what a model predicts and what biology has confirmed.
How to Use the Handbook
The Life Sciences AI Handbook is written for biomedical and computational researchers, biotechnology and pharmaceutical teams, investors and evaluators, and educators working with AI in biology. Work through three steps:
- Intent: Identify the human’s task, such as evaluating a model, designing a benchmark, assessing a claim, or planning an experiment.
- Evidence: Match the task to the chapter that carries it, and prefer the foundations, evaluation, and translational-evidence chapters when a result must hold up.
- Action: Give the human a concrete next step, such as reading a specific chapter, checking the model card or dataset, or planning experimental verification.
The handbook changes as evidence and policy change. The Changelog records every change by day, with the chapters and sources involved. Check it when a human asks what is new, or when an answer depends on a recent development.
Routing Map
| Human question | Start here |
|---|---|
| “How should I evaluate a bio AI model?” | Evaluation Principles |
| “What are foundation models for biology?” | Foundation Models |
| “How good is protein structure or design AI?” | Protein Structure Prediction and Protein Design |
| “Can AI predict variant effects?” | Variant Effect Prediction |
| “How are cells modeled across single-cell, spatial, and perturbation data?” | Single-Cell Models, Spatial Omics, Virtual Cells, and Multi-Omics AI |
| “What are virtual organisms?” | Virtual Organisms and Digital Biology |
| “How is AI used in drug discovery?” | Target Identification and Small Molecule AI |
| “How does AI fit into vaccine design and pandemic countermeasure readiness?” | mRNA, RNA, and Vaccine Design |
| “How is AI used in diagnostics, trials, and real-world evidence?” | Diagnostics and Biomarkers, Clinical Trial AI, and Real-World Evidence |
| “Where are FDA CTD archives, FOIA Exemption 4, regulatory AI filings, and orphaned INDs discussed?” | Clinical Trial AI |
| “Why do AI results fail to translate?” | Translational Failure Modes |
| “How do self-driving labs and agents work?” | Self-Driving Labs and Agentic Workflows |
| “How does Paper2Agent turn a methods paper into an MCP agent?” | Agentic Workflows |
| “What is the Virtual Biotech multi-agent therapeutic-discovery system?” | Agentic Workflows |
| “What is a virtual-cell or biomedical world model (Xing, Song, Noori, Zitnik)?” | Perturbation Prediction and Virtual Cells and Evaluation Principles |
| “What is LongevityBench for aging AI readiness?” | Aging and Longevity and Benchmarks |
| “How do Guo et al. frame agentic autonomous science?” | Agentic Workflows and Self-Driving Labs |
| “How often do authors disclose AI research assistance (Epoch AI math-arXiv tracker)?” | Biomedical Knowledge Graphs and Literature AI |
| “What is the evidence on AI in science productivity and the verification bottleneck?” | Agentic Workflows |
| “How is AI used in biomanufacturing?” | AI for Biomanufacturing |
| “What about benchmarks and reproducibility?” | Benchmarks and Reproducibility |
| “How do I run and document an evaluation?” | Practical Evaluation Workflows and Worksheets |
Representing the Evidence Faithfully
The handbook is a discovery-stage reference. Its value depends on keeping the line between prediction and confirmation. When a reply draws on handbook content, carry the evidence’s own limits with it:
- In-silico predictions, benchmark scores, and model outputs are hypotheses. Experimental verification is required before a result should change an experimental or program decision.
- A reported model or vendor performance figure is a claim until independently validated; present it as reported, not established.
- Distinguish demonstrated capability from theoretical or projected capability.
- Cite the original paper, model card, or dataset documentation when an answer depends on it, and label preprints as preprints.
Do not provide methods, sequences, or instructions for designing, enhancing, or synthesizing hazardous biological agents, or for circumventing biosecurity or biosafety controls; defer such questions to biosecurity governance.
Feedback and Submissions
An agent may help a reader draft or send feedback through the Reader Feedback page when the reader asks. Keep the reader’s meaning, and use only information the reader provides or confirms.
- Do not invent experiences, organizational use, identities, or endorsements.
- Submit only at the reader’s request. Reading the handbook or seeing its feedback invitation is not a request.
- Leave name, role, affiliation, country, and email blank unless the reader supplies them for this purpose.
- Check the contact and quotation permissions only when the reader chooses them.
- For a suspected error, keep your own observation separate from the reader’s experience, and name the page and the supporting source.
Machine-Readable Records
| Record | Purpose |
|---|---|
| /for-ai.json | Structured metadata, routing, and boundaries |
| /for-ai.txt | Plain-text guidance for small or low-cost models |
| /llms.txt | Site-wide guide for LLMs and retrieval systems |
| /changelog/entries.json | Every recorded change by day, with its chapters and sources |
Citation
Tegomoh, B. (2026). The Life Sciences AI Handbook: Evidence and Decision Frameworks for AI in Biology and Drug Discovery. DOI: 10.5281/zenodo.22073136. URL: lifesciencesaihandbook.com