From Reading Life to Reprogramming It
What is happening right now in AI, epigenetics, gene editing, regenerative medicine and biological research — and what it actually means for humans.
The short version
Biology is moving from a science that mainly observes living systems toward one that can increasingly read, predict, edit, reset, replace and design parts of them. AI is accelerating almost every step — but the laboratory, the clinic and the laws of biology still get the final vote.
CURRENT SNAPSHOT · 13 OCTOBER 2026
Written for curious non-specialists. Research claims are separated into peer-reviewed evidence, approved therapies, early human trials and company-reported results.
Why biology suddenly feels different
For most of modern history, biology was mainly a science of observation. Scientists could look at cells, classify diseases, measure molecules and slowly infer how living systems worked. The Human Genome Project made biology readable at a new scale. CRISPR made selected parts of DNA writable. Single-cell methods made it possible to examine individual cells rather than averaging entire tissues together. Now artificial intelligence is adding another layer: prediction.
The important shift is not that scientists have “solved life.” They have not. It is that several technologies that matured separately are now converging. Genome sequencing can tell us what DNA is present. Epigenomics can show how that DNA is being regulated. Single-cell tools can reveal which state a cell is in. Gene editing can alter selected sequences. Stem-cell methods can replace or rebuild some cell populations. AI can search patterns across all of those measurements at a scale no human researcher could inspect manually.
If our first Biology With Bun article asked what DNA and cells are, and the second asked how signals from behavior and the environment can travel downstream into cell biology, this third article asks a different question: what are scientists actually doing with that knowledge now?
THE NEW BIOLOGY STACK
READ biology → PREDICT biology → EDIT DNA → RESET cell state → REPLACE damaged cells → DESIGN new molecules → AUTOMATE experiments.
A vocabulary map before we start
You do not need a biology degree to follow the rest. Keep these four ideas in mind:
Genome: the complete DNA collection.
Epigenetics: regulation that affects how DNA is used without necessarily changing the DNA letters.
Cell state: the current molecular condition and behavior of a cell — what genes it is using, what proteins it is making and how it responds to signals.
AI model: a computational system trained to detect patterns and make predictions from large biological datasets.
1. READ: cells are becoming maps rather than mysteries
The first revolution is measurement. A tissue that once looked like “liver” or “brain” can now be broken into thousands or millions of individual cells and profiled one by one. Researchers can measure RNA, DNA methylation, chromatin accessibility, proteins and spatial location. This is revealing that tissues contain far more cell states than older textbooks suggested.
A major 2026 example came from the NIH Cellular Senescence Network. In June, researchers published the first large-scale atlas designed to identify and classify senescent cells across human tissues. Senescent cells have stopped dividing but remain metabolically active; they can help with wound healing and tumor suppression, yet they can also accumulate with age and contribute to inflammation and tissue dysfunction. The new atlas shows that “a senescent cell” is not one universal thing. Its identity depends on tissue, age and context. [1]
Another 2026 analysis combined more than 15,000 DNA-methylation profiles from 17 human tissues. It found both shared and tissue-specific patterns of epigenetic aging, including increasing molecular variability with age. This matters because it gives researchers a more precise map of what “older epigenetics” actually looks like across the body — and which parts might be more modifiable than others. [2]
Then, in October, gene-editing-based “cell recorders” were used to reconstruct the development of mouse embryos at extraordinary resolution, tracing how descendants of a single fertilized cell contributed to later tissues. That is developmental biology turning into lineage history: not simply what a cell is, but where it came from. [3]
WHY THIS MATTERS
Before biology can repair or reprogram cells, it needs to know what a healthy cell, an aged cell, a damaged cell and a developing cell actually look like at molecular resolution.
2. PREDICT: AI is learning to forecast what biology will do
Biology produces an awkward kind of data: enormous, noisy, interconnected and context-dependent. A single DNA change may alter RNA production, protein binding, cell behavior or nothing obvious at all. This is exactly the kind of high-dimensional pattern problem in which modern machine learning can be useful.
In September 2026, Google DeepMind launched the AlphaGenome Atlas, containing predictions for the molecular effects of roughly 9 billion possible single-letter DNA changes across the human genome. The atlas does not tell us whether every variant will cause disease. It predicts molecular consequences — for example how a change may affect gene regulation or other genomic processes — giving researchers a way to prioritize which variants deserve laboratory and clinical investigation. [4]
The larger ambition is the “virtual cell.” Biohub launched the Virtual Biology Initiative in April 2026, and on 7 October announced an expanded international effort totaling about $1.8 billion in funding, data, computing and measurement technology with partners including the U.S. Department of Energy and NIH. The goal is to generate the open, AI-ready biological data needed for predictive models of cells. [5]
A useful way to understand a virtual cell is not as a perfect digital human cell living inside a computer. We are nowhere near that. The near-term goal is narrower: a model that can increasingly predict how a particular cell may respond when a gene is changed, a drug is added, a signal arrives or a disease process develops.
THE DREAM — AND THE LIMIT
Instead of physically testing every possible intervention first, researchers want AI to narrow the search. But a prediction is still a hypothesis. Wet-lab experiments and clinical trials remain the reality check.
3. RESET: epigenetic rejuvenation has entered human testing
Epigenetics has become one of the most provocative areas in aging research because an old cell often still contains essentially the same DNA sequence it had when it was young. What has changed is partly the regulatory state: which genes are accessible, which programs are active and how stable the cell’s identity remains.
A 2026 review in Nature Reviews Molecular Cell Biology described aging as involving progressive loss of epigenetic fidelity — deterioration in the systems that keep gene-expression programs precise. Another Nature review published in September framed aging partly as erosion of the regulatory “grammar” that establishes and maintains cell identity during development. These are frameworks, not proof of one master cause of aging, but they capture a central idea: some age-related dysfunction may come from information-control problems as well as physical damage. [6][7]
The most dramatic translation of that idea is partial epigenetic reprogramming. Full reprogramming can push a mature cell back toward a pluripotent stem-cell state, which risks erasing cell identity and creating abnormal growth. Partial reprogramming instead tries to reset some age-associated gene-regulatory patterns while keeping the cell recognizably itself.
In June 2026, the first patient was dosed in a Phase 1 study of ER-100, an experimental therapy delivering controlled expression of three transcription factors — OCT4, SOX2 and KLF4, commonly abbreviated OSK — to retinal cells in people with optic neuropathies. On 8 October, the company reported the first interim human data: three glaucoma participants had tolerated the treatment through Day 56, and two showed preliminary improvement signals in visual-field testing. These results are far too early and too small to establish rejuvenation or efficacy; the important milestone is that controlled epigenetic restoration has moved from animal research into human safety testing. [8]
PUBLIC-READY TRANSLATION
We do not yet know how to make an older human body young again. What is now being tested is much more specific: can an aged or injured cell be nudged toward a younger functional state without changing its DNA sequence or losing its identity?
4. EDIT: gene editing is becoming a clinical toolkit, not one technology
CRISPR made targeted genome editing programmable, but 2026 shows how quickly the field is diversifying. Scientists now talk about nuclease editing, base editing and prime editing as distinct tools with different strengths.
CRISPR-based medicine is already real. In July 2026, the U.S. FDA expanded Casgevy — the first approved CRISPR-based therapy — to children as young as two with sickle-cell disease or transfusion-dependent beta-thalassemia. The treatment edits a patient’s blood-forming stem cells outside the body and returns them after conditioning. It is powerful medicine, but also logistically intensive and medically demanding. [9]
Base editing can change selected DNA letters without making the same kind of double-strand cut used by classic CRISPR-Cas9. In a Phase 1 study published in the New England Journal of Medicine in 2026, a single infusion of VERVE-102 produced dose-dependent, substantial and sustained reductions in PCSK9 and LDL cholesterol by editing liver cells in adults with inherited or premature cardiovascular disease. Safety and longer-term outcomes still require continued follow-up. [10]
Prime editing aims for even more flexible sequence correction. On 5 October 2026, Prime Medicine announced that the first patient had been dosed with PM577a, an in-vivo prime-editing therapy designed to correct a common ATP7B mutation causing Wilson disease. The Phase 1/2 trial is now underway, with initial clinical data expected in 2027. [11]
The field is also getting better at designing editors themselves. A 2026 Nature Biotechnology research highlight described AI-assisted creation of synthetic TnpB gene editors that do not exist in nature; selected variants showed stronger editing activity than natural versions in bacterial, plant and human cells while maintaining comparable specificity. [12]
5. REPLACE: regenerative medicine is crossing into ordinary clinical reality
If gene editing changes instructions, regenerative medicine changes the cells doing the work. Induced pluripotent stem cells, or iPSCs, are mature cells reprogrammed into a stem-cell-like state and then directed into another specialized cell type. The concept is now moving beyond proof-of-principle.
In March 2026, Japan granted conditional and time-limited marketing approval to the world’s first two therapies developed from iPSCs: one for Parkinson’s disease and one for heart failure. The Parkinson’s therapy uses iPSC-derived dopaminergic cells intended to replace neurons lost in the disease. Conditional approval is not the same as final proof of long-term benefit, but it marks a major step from laboratory reprogramming toward regulated medicine. [13]
A separate 2026 Phase 1/2 study published in Nature Medicine transplanted human pluripotent-stem-cell-derived dopaminergic progenitors into the brains of eight people with Parkinson’s disease. Seven completed twelve-month follow-up; imaging supported graft survival, and no tumor formation was seen. One participant died from a severe fungal infection during immunosuppression, underscoring that cell replacement brings risks beyond the cells themselves. [14]
Regeneration research is also revealing that adult cells can be more flexible than once assumed. In October 2026, Nature reported that alveolar stem cells in the lung could transdifferentiate — change identity — to drive bronchiolar regeneration in experimental models. This does not mean humans can regrow an arm. It does show that adult repair sometimes involves cells crossing identity boundaries when the right injury signals and regulatory programs are present. [15]
WHAT REGENERATION IS BECOMING
Not one miracle therapy. A toolbox: replace lost cells, engineer tissues, redirect cell identity, control scarring, restore blood supply and learn which developmental programs can be safely reopened.
6. DESIGN: AI is beginning to create biological parts that evolution never made
The first wave of AI in biology predicted structures. The newer wave increasingly designs sequences.
Generative protein models can now propose proteins with chosen shapes, binding properties or functions. A July 2026 Nature paper showed AI-redesigned starting proteins could improve experimental evolution, creating proteins that were more robust and easier to evolve toward new functions. Other 2026 work demonstrated zero-shot design of proteins that bind small-molecule drugs and AI frameworks that can shrink or modify natural proteins while preserving key structural features. [16][17]
Synthetic biology is moving toward equally strange territory. In September 2026, a Nature paper described RNA transfer vehicles built from AI-designed synthetic protein assemblies rather than copied directly from natural viruses. The goal is not to create artificial life; it is to engineer biological delivery machinery with properties selected for a task. [18]
This is one of the clearest places where AI changes the search process. Natural evolution explores possible proteins over enormous timescales. AI can suggest candidate sequences in silico, after which laboratory screening determines which designs actually fold and function. The machine proposes. Biology decides.
SMART HUMOR, SERIOUS POINT
AI can design a protein before lunch. The protein still has to survive reality after lunch.
7. DISCOVER: AI is moving from analyst to laboratory collaborator
A particularly important transition is happening in the scientific workflow itself. AI used to sit mostly at the analysis stage: researchers generated data, then models helped interpret it. Increasingly, AI is being used earlier — to generate hypotheses, choose experiments and propose molecules — and automation is carrying some of those choices into the laboratory.
On 23 September 2026, Anthropic announced a life-sciences research group and wet lab and reported that Claude had searched DNA datasets, proposed hypotheses and helped identify an uncharacterized enzyme system with CRISPR-like repeats, which scientists then tested experimentally. This is a company-reported early result rather than an independently established new field standard, but it illustrates the direction clearly: AI participating in the cycle of observation → hypothesis → experiment. [19]
At the same time, “self-driving laboratories” are moving from academic demonstrations toward industrial use. A 2026 Nature Reviews Chemistry review described systems that combine robotics, automated experimentation and AI so algorithms can propose, execute and interpret experiments with limited human intervention. An October Nature Synthesis issue focused specifically on the difficulties of making such laboratories scalable, standardized and reliable in industry. [20][21]
The most consequential version of this is not a robot scientist replacing every biologist. It is a research loop that can run far faster: model proposes → robot performs → instruments measure → model updates → next experiment begins.
WHY THIS COULD MATTER MORE THAN CHATBOTS
Scientific progress is partly limited by how many experiments humans can design, run and interpret. Automating parts of that loop can increase the number of hypotheses tested — provided the experiments remain trustworthy and well designed.
8. MEDICINE: AI-designed molecules are starting to face the only test that matters — patients
Drug discovery is one of the clearest commercial uses of AI biology, but it is also where hype has to confront clinical evidence.
A generative-AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis reached a randomized Phase 2a trial, published in Nature Medicine in 2025 — important because it moved an AI-generated drug program beyond early human safety testing. In October 2026, Iambic Therapeutics submitted an FDA investigational-new-drug application for IAM217, an AI-assisted KIF18A inhibitor intended for trials in solid tumors. That submission is a company milestone, not evidence the drug works in patients yet. [22][23]
AI is also being used to design antibodies and improve clinical-trial operations. A blinded 2026 benchmark in Nature Biotechnology experimentally tested more than 500 AI-designed or AI-ranked antibodies from 29 organizations and found that several groups produced highly potent, developable candidates. Meanwhile, researchers are evaluating AI for trial recruitment, monitoring and decision support — areas where real-world evidence has already shown that useful-sounding AI does not automatically improve patient outcomes. [24][25]
9. What is still not solved
The speed of the field makes it easy to confuse a technological milestone with a medical solution. They are not the same thing.
Prediction is not causation. A model can identify a molecular pattern without proving what drives it.
A younger epigenetic clock is not automatically a younger, healthier organism.
Editing the correct DNA sequence is useless if the editor cannot safely reach the correct cells.
Cell replacement can create new problems: immune rejection, immunosuppression, abnormal growth and integration into existing tissue.
AI-designed molecules still fail for ordinary biological reasons: toxicity, metabolism, delivery, poor efficacy or unexpected interactions.
Virtual cells are partial models of biological behavior, not complete digital copies of living humans.
Aging is multi-system. Resetting one hallmark does not prove lifespan or healthspan will meaningfully increase.
Biological data can be biased, incomplete and inconsistent. Models trained on it inherit those limitations.
The most mature research programs therefore combine prediction with experiment. The strongest claim an AI model can make is usually not “this will work,” but “this is a better place to look.”
10. What this actually means for humans
The most important change is conceptual. Medicine has traditionally intervened after a system has gone wrong: lower the blood pressure, kill the infection, replace the missing hormone, remove the tumor. That will remain essential. But modern biology is increasingly asking whether it can intervene closer to the underlying state of the cell.
That can mean correcting a DNA variant, changing gene regulation, replacing a lost cell type, restoring a damaged tissue, designing a protein that does not exist in nature, or predicting how a cell will react before giving a drug to a patient. None of these technologies gives us complete control over biology. Together, they change what “possible medicine” looks like.
The phrase “programmable biology” should therefore be used carefully. Living systems are not laptops. Cells adapt, mutate, communicate, age and respond to context. But parts of biology are becoming more engineerable because we can increasingly measure the state, identify the control points and intervene with precision.
THE CLEAREST 2026 SUMMARY
Biology is becoming more readable, more predictable and more editable. Epigenetics is making cell state a therapeutic target. Regenerative medicine is replacing selected cell populations. AI is compressing the search space. The result is not immortality. It is a rapidly expanding ability to ask biology more precise questions — and sometimes to change the answer.
2026: a compressed timeline
March
Japan grants conditional approval to the first two iPSC-derived regenerative medicines.
May–August
VERVE-102 base-editing human data are published in NEJM, showing sustained PCSK9 and LDL reductions after one infusion.
June
NIH-backed researchers publish the first large-scale human senescent-cell atlas; ER-100 doses the first participant in a human epigenetic-restoration trial.
June–July
Large epigenetic-aging analyses and reviews sharpen the idea that aging involves loss of regulatory fidelity as well as damage.
July
FDA expands the CRISPR therapy Casgevy to children as young as two for sickle-cell disease and beta-thalassemia.
September
DeepMind releases AlphaGenome Atlas; AI-designed RNA transport assemblies are published; Anthropic launches a life-sciences wet lab.
October 5
The first patient is dosed with in-vivo prime editor PM577a for Wilson disease.
October 7
Biohub and partners expand the Virtual Biology Initiative to a roughly $1.8 billion coordinated effort.
October 8
Life Biosciences reports the first interim human data from ER-100; Nature reports cell-recording maps of embryo development.
Source notes & further reading
This article prioritizes peer-reviewed research, government/regulatory sources and primary institutional announcements. Company announcements are labeled as such in the text. Links were checked during research on 13 October 2026. Inclusion does not imply endorsement.
[1] NIH — first large-scale atlas of senescent cells across the human body (11 June 2026) — Source
[2] Nature Aging — Meta-analysis of DNA methylation aging signatures in 17 human tissues (26 June 2026) — Source
[3] Nature News — gene-editing cell recorders map embryo development (8 October 2026) — Source
[4] Google DeepMind — AlphaGenome Atlas, predictions for 9 billion single-nucleotide variants (8 September 2026) — Source
[5] Biohub — Virtual Biology Initiative expansion, $1.8 billion coordinated commitment (7 October 2026) — Source
[6] Nature Reviews Molecular Cell Biology — Systemic epigenetic dysregulation as a driver of ageing (27 March 2026) — Source
[7] Nature — Written in development and lost in ageing: the grammar of cellular identity (30 September 2026) — Source
[8] Life Biosciences — first-in-human interim ER-100 data (company announcement, 8 October 2026) — Source
[9] U.S. FDA — Casgevy expanded to patients aged 2 years and older (1 July 2026) — Source
[10] New England Journal of Medicine — In vivo base editing with VERVE-102 (published 25 May 2026) — Source
[11] Prime Medicine — first patient dosed with PM577a in-vivo prime editor for Wilson disease (company announcement, 5 October 2026) — Source
[12] Nature Biotechnology — AI-assisted design of synthetic gene editors (11 August 2026) — Source
[13] Nature Biotechnology — world’s first two iPSC therapies in Japan (17 April 2026) — Source
[14] Nature Medicine — stem-cell-derived dopaminergic cells for Parkinson’s disease, Phase 1/2 (9 July 2026) — Source
[15] Nature — alveolar stem cells transdifferentiate to drive bronchiolar regeneration (7 October 2026) — Source
[16] Nature — AI-redesigned starting points and outcomes enhance protein evolution (22 July 2026) — Source
[17] Nature — Zero-shot design of drug-binding proteins (24 June 2026) — Source
[18] Nature — AI-designed synthetic protein assemblies used as RNA transfer vehicles (2 September 2026) — Source
[19] Anthropic — Claude discovers a novel enzyme system with CRISPR-like repeats (company report, 23 September 2026) — Source
[20] Nature Reviews Chemistry — The past, present and future of self-driving laboratories (31 July 2026) — Source
[21] Nature Synthesis — The road to self-driving laboratories in industry (8 October 2026) — Source
[22] Nature Medicine — generative-AI-discovered TNIK inhibitor, randomized Phase 2a trial (2025) — Source
[23] Iambic Therapeutics — IND submission for AI-assisted KIF18A inhibitor IAM217 (company announcement, 5 October 2026) — Source
[24] Nature Biotechnology — blinded prospective benchmark of AI antibody discovery (19 August 2026) — Source
[25] Nature Medicine — pragmatic randomized trial of generative-AI clinical decision support (26 June 2026) — Source
CLOSING NOTE
The frontier is not one technology. It is the convergence of better biological measurement, programmable gene tools, cell reprogramming, regenerative medicine, automation and AI. The responsible question is not only “Can we change biology?” but “What exactly are we changing, how do we know it worked, and what risks travel with the intervention?”
THE NEW BIOLOGY
BunCosmos AI Research Lab · Biology With Bun · Research edition · October 2026