Executive Summary

Life sciences AI now reaches beyond isolated modelling tasks into biological discovery, translational research, and research operations. Protein structures are predicted before they are solved. Cells are represented by foundation models. Organismal and environmental systems are becoming model targets. Molecules are generated faster than they can be made at the bench. Autonomous chemistry runs closed loops on cloud robotic platforms. The practical risk is not only overclaiming. It is using the wrong evidence for the wrong decision.

Core Takeaways

The handbook organises evidence into three tiers throughout. The summary below names the bottom-line state of each major area as of May 2026.

Protein structure prediction is a routine research input. AlphaFold 2 effectively solved single-chain structure prediction for the majority of well-folded protein domains at CASP14. AlphaFold 3 extended prediction to biomolecular interactions involving nucleic acids, ions, and small-molecule ligands. The initial AlphaFold 3 release was restricted; the open-source response (Boltz, Chai) reproduced AF3-class capability with permissive licenses. Structure prediction is now routine research infrastructure. Confidence-metric discipline (pLDDT, PAE, ipTM) and validation-experiment planning are the value-adding skills.

Protein design is an experimental discipline supported by generative models. RFdiffusion generates backbones. ProteinMPNN and LigandMPNN design sequences. ESM3, Chroma, and ProGen2 add multimodal and autoregressive generative approaches. RFantibody (Nature 2026) extends the lineage to peer-reviewed de novo antibody design. AlphaProteo from DeepMind reports strong binder generation but remains an unpublished arXiv preprint with restricted code. Every designed protein still needs expression, purification, characterisation, functional assay, and developability review before it counts as a candidate.

Single-cell foundation models do part of what their papers claim. scGPT, Geneformer, scFoundation, and scBERT produce useful general-purpose cell-state representations. They support cell-atlas search (SCimilarity, Nature 2025) and integration tasks competitively. Nicheformer (Nature Methods 2025) is the strongest current peer-reviewed signal that spatial context belongs in the pretraining corpus rather than only in downstream analysis. Independent evaluations published in 2024 and 2025 (Ahlmann-Eltze et al., Nature Methods; Boiarsky et al., Nature Machine Intelligence) show that on perturbation prediction and several core tasks, the deep approaches do not consistently outperform PCA plus linear regression. The evidence supports scFMs for some tasks and leaves others unverified.

Genome foundation models are early and active. Evo (Science 2024) demonstrated a 7-billion-parameter model reasoning across DNA, RNA, and protein. Evo 2 (Nature 2026) extended to 40 billion parameters across all domains of life, with open model parameters, code, and OpenGenome2 data. Nucleotide Transformer (Nature Methods 2025), GET (Nature 2025), Orthrus (Nature Methods 2026), and AlphaGenome (Nature 2026) show the field splitting into genome-wide, transcriptional, RNA-specific, and regulatory-variant models. Production-quality variant interpretation still happens inside the ACMG/AMP framework with AI predictors as one evidence layer, not as classifications.

Systems biology remains the weak link. Cell-state embeddings, pathway scores, and network diagrams become useful only when they name a mechanism and a falsifying perturbation. Gene regulatory network inference and whole-cell modeling are real research traditions, but observational edges are not causal maps. Virtual-organism claims are not larger virtual-cell claims; they require development, physiology, tissue coupling, environment, time, and prospective validation.

Organismal and environmental biology belong in the core scope. Neuroscience AI, aging clocks, plant and crop AI, ecological monitoring, environmental DNA, and virtual organisms are not side cases. They are the scale rung between cellular biology and translation. The evidence problem is measurement validity: whether the model is learning biology or platform artifacts, geography, sampling effort, season, field conditions, or cohort structure.

AI for therapeutics is delivering candidates, not yet approvals. Insilico Medicine’s ISM001-055 (rentosertib) reached randomized Phase 2a testing with peer-reviewed Nature Medicine evidence after the discovery and Phase I narrative appeared in Nature Biotechnology. Recursion has multiple AI-discovered candidates in clinical trials. No AI-discovered small molecule has been approved by FDA or EMA. The clinical attrition rate for AI-discovered candidates so far looks similar to traditionally discovered ones at the same stage; the AI advantage is in cycle time and breadth, not in pivotal success rates.

Translation is broader than small molecules. Chemical biology, target engagement, cell and gene therapy, diagnostics, biomarkers, trials, and real-world evidence each have different validation objects. A generated molecule, engineered cell, biomarker signature, or diagnostic feature becomes useful only when the evidence matches the decision it is supposed to support.

Self-driving laboratories and agentic systems work for bounded tasks. Coscientist (Nature 2023) demonstrated LLM-planned Pd-catalysed coupling. Virtual Lab (Nature 2025) demonstrated multi-agent biology research producing experimentally validated SARS-CoV-2 nanobodies. A-Lab (Nature 2023) reported 41 new inorganic materials and drew matters-arising critiques that made the validation discipline unavoidable. ChemCrow (Nature Machine Intelligence 2024) wraps LLMs around chemistry-specific tools. Open-ended autonomous biology beyond bounded settings remains beyond current capabilities.

Information hazards are handled through deliberate disclosure. The standard is not secrecy by default. The standard is enough detail for scientific verification without unnecessary operational detail that raises misuse risk. The 2024 NIH and HHS DURC framework provides institutional context. The AlphaFold 3 restricted-release and Boltz/Chai open-source response is the live case study.

Reproducibility is computational and experimental. The DOME framework (Walsh et al., Nature Methods 2021) is the per-paper reporting standard. Model cards and dataset cards are the documentation layer. Bridge2AI and ARPA-H IGoR are the federal infrastructure programs. A notebook that reruns is necessary but not sufficient if the assay cannot be repeated.

Workforce and compute are institutional, not individual. A credible team is cross-disciplinary: biological expertise, data engineering, ML, wet-lab partnership, regulatory engagement when relevant, and governance. Compute without experimental judgement creates expensive noise.

Reading Rule

Every major capability claim in the handbook lands in one of three tiers:

• Demonstrated for claims supported by peer-reviewed evidence, official documentation, or reproducible blind benchmarks.
• Theoretical for claims plausible under current methods but not yet established for routine use.
• Beyond current capabilities for claims not supported by credible evidence with current systems.

If a claim cannot be placed in a tier, the claim is not specific enough.

Practical Use

The handbook is built to be read in the situations that matter:

• Evaluating a model claim before adopting it in a research program.
• Designing a validation plan that matches the biological decision the model informs.
• Briefing a research team on the current evidence for a given model class.
• Reviewing a vendor pitch against the published evidence in peer-reviewed venues.
• Deciding whether a newly published paper changes a program decision.

It is not a tutorial in machine learning. It is a reference for working biologists, biotechnology leaders, physician-scientists, drug-discovery scientists, synthetic biologists, plant biologists, ecologists, neuroscientists, aging researchers, graduate students, and research program leaders who need to evaluate AI systems against experimental evidence.

The One-Page View

The capability frontier is real and uneven. Structure prediction and selected design tasks are demonstrated. Single-cell representation works; single-cell perturbation prediction is contested. Genome foundation models are early. Systems biology and virtual-organism claims require stronger perturbation evidence. Plant, ecological, neuroscience, and aging AI are now large enough to deserve explicit homes. AI for therapeutics is delivering candidates into trials but not yet approvals. Autonomous laboratories work for bounded tasks. Information hazards require institutional discipline. Reproducibility requires infrastructure investment.

The value in life sciences AI is not the model alone. It is the validation plan, the wet-lab follow-up, the calibration discipline, and the willingness to treat every model output as a hypothesis with explicit uncertainty. Programs that internalise this read the literature differently and produce more reliable science.