Medicine & Longevity · Special Report
A technology once dismissed as too fragile and too risky has become the most transformative platform in modern medicine. Here is how it happened — and what comes next.
Nature Medicine · NEJM · BioNTech · Moderna · NIH · FDA · DeepMind AlphaFold
The story of mRNA medicine begins not with a pandemic but with a persistent, underfunded researcher and a molecule that everyone else had given up on. Katalin Karikó spent most of the 1990s and 2000s trying to convince her colleagues and grant-review committees at the University of Pennsylvania that messenger RNA — a short-lived molecular instruction slip that tells cells which proteins to build — could form the basis of a new class of medicine. She was demoted. Her grants were repeatedly denied. The scientific consensus held that mRNA was too unstable, too immunogenic, and too technically difficult to ever be a reliable drug platform.
She was right. Everyone else was wrong.
The discovery that Karikó made with immunologist Drew Weissman — that a specific chemical modification to mRNA's building blocks (replacing uridine with pseudouridine) could render it invisible to the immune system's alarm response — was the key that unlocked the entire field. It was published in 2005. It was awarded the Nobel Prize in Physiology or Medicine in 2023. In between, it became the scientific foundation for the fastest vaccine development campaign in history and the opening move in what many scientists now describe as a complete reimagining of medicine.
The COVID-19 vaccines developed by BioNTech/Pfizer and Moderna were the public's dramatic introduction to this technology. But they were never the destination. They were proof of concept on an unprecedented scale — 13 billion doses administered, extraordinary efficacy data generated, and a global manufacturing and regulatory pathway established in real time. What comes after COVID is where the revolution truly begins: personalised cancer vaccines, infectious disease protection for diseases that have resisted conventional vaccine approaches for decades, therapies for rare genetic disorders, and — in the more distant future — the possibility of turning off the molecular switches that drive ageing.
Threading through all of it is artificial intelligence. AI has not merely accelerated mRNA medicine; it has fundamentally changed what is possible within it. From protein structure prediction to mRNA sequence optimisation, from clinical trial design to personalised neoantigen identification, AI tools are collapsing timelines that once measured in decades to ones that measure in months.
Every cell in your body contains the same DNA blueprint. What makes a liver cell different from a neuron is which parts of that blueprint are being actively read at any given moment. mRNA is the intermediary — a temporary transcript copied from a specific gene that carries instructions out of the nucleus to the ribosomes, the cell's protein-building machinery. Once the ribosome reads the mRNA and builds the specified protein, the mRNA degrades. It never enters the nucleus, cannot alter DNA, and leaves no permanent record.
Therapeutic mRNA exploits this natural process. Instead of injecting a protein directly into the body — which is what conventional vaccines and many drugs do — mRNA medicine delivers the instructions for cells to build that protein themselves. This offers several profound advantages: the body produces the protein in its natural three-dimensional form, the immune response is often stronger and more durable, and the same manufacturing platform can be rapidly reprogrammed to produce instructions for a different protein with minimal changes to the production process.
The critical technical hurdle was the immune system's tendency to attack synthetic mRNA as a foreign invader, triggering dangerous inflammation. Karikó and Weissman discovered that replacing the natural nucleoside uridine with pseudouridine — a chemically modified version found in some natural RNAs — made synthetic mRNA effectively invisible to the toll-like receptors that normally detect and destroy it. This single modification, protected by a patent that BioNTech licensed in 2013, is foundational to every approved mRNA medicine.
On January 11, 2020, Chinese scientists published the genetic sequence of a novel coronavirus. On January 13, 2020 — two days later — Moderna's team had already designed the mRNA sequence for what would become mRNA-1273. That single data point captures why mRNA medicine is so radical: the entire molecular design process, which takes years for a conventional vaccine, was completed in 48 hours. The subsequent 11 months were spent manufacturing, testing, running clinical trials, and navigating regulatory approval — not designing the molecule.
Both Moderna and BioNTech/Pfizer independently chose to target the same thing: the SARS-CoV-2 spike protein, the distinctive crown-shaped protrusion the virus uses to enter cells. They also both chose to encode the spike in its prefusion-stabilised form — a structural detail that turned out to be crucial for eliciting a strong immune response, and one that drew on years of prior work by Barney Graham and Jason McLellan at the NIH on coronavirus structural biology.
The phase 3 trial results, published in December 2020, were remarkable: 94.1% efficacy for Moderna's vaccine and 95% for Pfizer/BioNTech's, at a time when regulators had set a minimum acceptable bar of 50%. Both vaccines received Emergency Use Authorisation within days of each other. By the end of 2021, the global mRNA vaccine campaign had prevented an estimated 14–20 million deaths in its first year alone, according to analyses published in The Lancet.
Beyond the efficacy numbers, the COVID campaign provided something invaluable to the field: a vast real-world dataset on mRNA safety and immune biology at population scale. Over 13 billion doses administered across diverse populations, ages, and health conditions generated a pharmacovigilance dataset that would have taken decades to accumulate in normal circumstances. The signal on myocarditis in young males following Moderna's higher-dose formulation, for example, was identified, characterised, and incorporated into clinical guidelines within months — a surveillance speed that conventional post-marketing systems have never matched.
As SARS-CoV-2 evolved, producing Alpha, Delta, Omicron, and subsequent variants, the mRNA platform demonstrated another of its core advantages: speed of iteration. Moderna produced an updated Omicron-targeting mRNA sequence within 30 days of the variant being identified. The bivalent boosters targeting both the original strain and Omicron sub-variants reached regulators for approval in 2022, bypassing the need for a full new clinical trial programme because the platform's safety and the regulatory process for updating it had already been established.
This iterability will become critically important as the platform moves into diseases where the pathogen evolves rapidly — influenza, HIV, and respiratory syncytial virus (RSV) — or where the target antigen needs to be personalised per patient, as in cancer.
The COVID vaccines didn't just save millions of lives. They were the largest and fastest clinical validation in the history of medicine — for a technology that is now going to be applied to nearly every disease on the planet.
— Ugur Sahin, Co-founder & CEO, BioNTech
For the founders of BioNTech, COVID was always the detour. Ugur Sahin and Özlem Türeci founded the company in 2008 with a single goal: personalised cancer immunotherapy using mRNA. The idea had been dismissed as wildly ambitious — not because the biology was wrong, but because the technical and economic barriers to making a truly personalised cancer medicine at scale seemed insurmountable. Each patient's cancer carries a unique set of mutations. Each personalised vaccine would need to be designed, manufactured, quality-controlled, and administered on a patient-by-patient basis, within a window of weeks to months before the cancer progresses. That requires a speed and precision in molecular medicine that simply did not exist before AI.
Here is why cancer vaccination with mRNA is different from vaccination against an infectious disease. When a cell becomes cancerous, it acquires mutations — changes to its DNA — that cause it to produce altered proteins not found in healthy cells. These altered proteins are called neoantigens. The immune system can in principle recognise neoantigens as foreign and attack cells displaying them. The problem is that most cancers evolve mechanisms to hide from or suppress the immune system. A personalised mRNA cancer vaccine works by taking a biopsy of a patient's tumour, sequencing its DNA, identifying the neoantigens unique to that cancer, and then synthesising an mRNA encoding those neoantigens — essentially training the patient's immune system to recognise and destroy their specific cancer.
The first major clinical signal came in 2023, when Moderna and Merck reported phase 2b trial results for mRNA-4157/V940 — a personalised neoantigen vaccine combined with the checkpoint inhibitor pembrolizumab (Keytruda) — in patients with high-risk stage III/IV melanoma who had undergone surgery. The combination reduced the risk of cancer recurrence or death by 49% compared to pembrolizumab alone. This was the first randomised controlled trial to demonstrate that a personalised mRNA neoantigen vaccine meaningfully improves outcomes in cancer patients, and it triggered a global surge of interest and investment in the space.
Without AI, personalised cancer vaccines are not practically feasible at scale. The process requires sequencing a patient's tumour, comparing it to their healthy genome to identify somatic mutations, predicting which of those mutations produce neoantigens that will be displayed on the cell surface (a process called MHC presentation prediction), selecting the best subset to include in the vaccine, and optimising the mRNA sequence — all within a timeframe compatible with clinical care, typically 4–8 weeks from biopsy to first injection.
AI tools now handle much of this pipeline. Deep learning models predict with high accuracy which mutant peptides will bind to a given patient's MHC molecules — the cell-surface proteins that present antigens to the immune system. This step, previously requiring laborious laboratory binding assays, can now be completed computationally in hours. Moderna's mRNA-4157 pipeline uses AI-driven neoantigen selection as its core engine; BioNTech's BNT111, BNT112, and BNT122 programmes all incorporate deep learning at multiple stages of target identification and sequence design.
AlphaFold, DeepMind's protein structure prediction model — arguably the single most consequential AI system deployed in life sciences — has further transformed the field by making it possible to predict how a target protein folds in three dimensions, and therefore how a candidate mRNA-encoded antigen will be presented and recognised. The structure prediction problem that AlphaFold solved had been considered one of biology's hardest open questions for 50 years. Its solution in 2020 was not a marginal improvement; it achieved accuracy comparable to experimental crystallography in a fraction of the time and cost.
The intersection of AI and mRNA medicine is not a single technology — it is a cascade of tools operating at every layer of the drug development stack, from the atomic level of protein folding to the population level of clinical trial design. Understanding each layer clarifies both how far the field has come and how much further it can go.
| AI Application | What It Does | Key Tool / Platform | Impact on Timeline | Stage |
|---|---|---|---|---|
| Protein Structure Prediction | Predicts 3D folding of target proteins and antigens with near-crystallographic accuracy | DeepMind AlphaFold 2 & 3 | Weeks → hours; cost reduction ~1,000× | Production |
| Neoantigen Prediction | Identifies tumour-specific mutations likely to elicit immune response from sequencing data | Moderna AI pipeline, BioNTech iNeST, pVACtools | Months → 4–6 weeks per patient | Phase 2/3 |
| mRNA Sequence Optimisation | Optimises codon usage, UTR design, and secondary structure for stability and expression | Moderna mRNA Studio, LinearDesign (ByteDance) | Reduces degradation; increases protein yield 3–10× | Production |
| Lipid Nanoparticle Design | Designs LNP formulations for targeted organ delivery (liver, lung, tumour, lymph node) | Insilico Medicine, Absci, MIT AI labs | Enables tissue-specific targeting not previously feasible | Phase 1/2 |
| MHC Binding Prediction | Predicts which peptides will be displayed on a patient's specific HLA molecules | NetMHCpan, DeepHLApan, MHCflurry | Replaces binding assays; near-instant per patient | Production |
| Clinical Trial Design | Adaptive trial designs, patient stratification, predictive biomarker identification | Tempus, Flatiron, various pharma AI platforms | Reduces required sample sizes; accelerates signals | Production |
| Generative Protein Design | Designs entirely novel proteins not found in nature for enhanced immunogenicity | RFdiffusion, ProteinMPNN (Baker Lab) | Expanding design space beyond natural sequences | Early Research |
One of the more unexpected contributions to mRNA optimisation came from researchers at ByteDance — the company behind TikTok. Their LinearDesign algorithm, published in Nature in 2023, frames mRNA codon optimisation as a problem structurally similar to natural language sequence optimisation. By adapting techniques from computational linguistics, LinearDesign can identify mRNA sequences that produce dramatically higher protein yields while maintaining stability — in some cases increasing expression 3–10× compared to standard sequences. It is now widely used in both vaccine and therapeutic mRNA development, and represents a template for how AI tools developed for entirely different domains are finding unexpected applications in biology.
The COVID vaccines were the first, but the pipeline behind them has grown dramatically. As of early 2025, over 100 mRNA-based therapeutic candidates are in active clinical development across oncology, infectious disease, rare genetic disorders, and autoimmune disease. The following represents the most clinically advanced and scientifically significant programmes.
The excitement around mRNA is justified — but the field has a history of over-promising that its current generation of leaders is trying to correct. A clear-eyed account of the remaining challenges is essential to understanding the realistic trajectory.
The COVID vaccines were administered intramuscularly, targeting immune cells in the draining lymph nodes — a relatively forgiving delivery target. Many of mRNA medicine's most important future applications require delivery to specific organs: the liver for metabolic disease, the lung for cystic fibrosis, tumour tissue for cancer immunotherapy, and the brain for neurological disorders. Current LNP technology delivers primarily to the liver due to the organ's natural role in clearing blood-borne particles. Achieving reliable targeted delivery to other tissues remains an active research challenge, with AI-assisted LNP design being one of the most promising approaches.
The original Pfizer/BioNTech COVID vaccine required storage at −70°C — a logistical barrier that effectively excluded large parts of the developing world from the rollout. Subsequent formulations have been improved to standard refrigerator temperatures (2–8°C), and further stability work is ongoing. For mRNA to fulfil its potential as a global health tool, room-temperature-stable formulations will be necessary, particularly for applications in lower-resource settings. Lyophilisation (freeze-drying) and novel excipient chemistries are being actively explored.
mRNA's intentional transience — the fact that it degrades within days — means that for therapeutic applications requiring sustained protein expression (such as enzyme replacement therapy or certain gene therapy applications), repeated dosing is necessary. This is clinically manageable but adds cost and patient burden. Self-amplifying mRNA (saRNA), which encodes its own replication machinery and therefore requires a much lower initial dose for sustained expression, is in early clinical development and may address this limitation.
The 49% efficacy signal from the personalised melanoma vaccine trial is remarkable — but manufacturing a bespoke mRNA vaccine for each individual patient, within a clinically meaningful timeframe and at a cost compatible with healthcare system adoption, remains a profound challenge. Moderna's manufacturing process for mRNA-4157 has compressed the biopsy-to-vaccine timeline to approximately 6–8 weeks. Scaling this to hundreds of thousands of cancer patients globally will require automation and infrastructure investment that is still being built.
Perhaps the most underappreciated challenge in mRNA medicine is not scientific but structural. The personalised cancer vaccines that represent the technology's most transformative potential will, at least initially, be extremely expensive — potentially $100,000–$500,000 per patient course for a personalised neoantigen vaccine. The question of who has access to this medicine, and how health systems in middle- and lower-income countries can participate in what may be the most consequential medical platform of the 21st century, is one that the field has not yet adequately addressed. The COVID vaccine rollout's equity failures — where high-income countries administered doses while low-income countries waited — are an uncomfortable precedent.
The cancer vaccine pipeline dominates the current conversation, but researchers are actively pursuing mRNA applications in several other domains that could prove equally transformative.
HIV has defeated every vaccine strategy attempted since the virus was identified in 1983. Its extreme genetic variability — far greater than SARS-CoV-2 — has meant that a neutralising antibody response sufficient to block infection has never been achieved with conventional approaches. mRNA's speed of update and the ability to encode mosaic antigens covering broad genetic diversity make it the most credible current approach. The IAVI/Moderna programme is specifically designed around an mRNA prime that includes a germline-targeting immunogen — an antigen engineered to stimulate the specific rare B cells that can evolve into broadly neutralising antibody producers. Phase 2 data is anticipated by 2026.
TB kills approximately 1.3 million people annually — more than any other single infectious agent — and the BCG vaccine, despite being over 100 years old, offers only partial and highly variable protection against pulmonary disease in adults. Malaria kills over 600,000 people per year, most of them children under five in sub-Saharan Africa. Both represent high-priority targets for mRNA vaccine development, with multiple programmes in early-stage clinical testing as of 2025. The economics here are challenging — neither disease offers the commercial incentives that drove COVID vaccine development — and sustained public and philanthropic funding will be essential.
CRISPR-based gene editing requires delivery of two components: the Cas9 protein and a guide RNA. mRNA encoding Cas9 — rather than a DNA vector — offers a transient, non-integrating approach to delivering the editing machinery that avoids many of the safety concerns associated with viral vectors. Intellia Therapeutics and others are already in clinical trials using mRNA-LNP systems to deliver CRISPR components for genetic liver disease. This represents mRNA operating not as a therapy itself but as the delivery vehicle for an entirely different class of medicine.
The most speculative but scientifically intriguing mRNA applications concern cellular reprogramming and longevity. Research groups at Harvard and the Salk Institute have demonstrated that episodic expression of Yamanaka factors — the transcription factors that can reset cells to a pluripotent stem-cell-like state — can reverse cellular hallmarks of ageing in mouse models without causing cancer. mRNA delivery of these factors allows for precise, transient expression that can be titrated and stopped. Human trials remain years away, but the conceptual connection between mRNA's inherent transience and its suitability for controlled, reversible cellular reprogramming is one that the longevity research community is taking increasingly seriously.
We are at the beginning of a new era in medicine. The question is no longer whether mRNA can work. The question is how fast we can build the manufacturing, clinical, and regulatory infrastructure to deploy it everywhere it is needed.
— Stéphane Bancel, CEO, Moderna
What distinguishes mRNA medicine from every previous pharmaceutical revolution is the concept of platform generalisability. The manufacturing process, the delivery chemistry, the regulatory framework, the clinical infrastructure, and the AI tools developed for COVID vaccines are reusable and reprogrammable for every subsequent mRNA application. Each new product is not built from scratch; it is a new instruction set running on an established platform.
This is a structural shift in how medicine is produced. The pharmaceutical industry's traditional model — decades of development, billions of dollars, a product designed for a specific disease — is being supplemented (and for certain disease categories, replaced) by a model that looks more like software development: a shared platform, rapid iteration, modular components, and AI-assisted design operating on timescales of weeks rather than years.
The implications extend well beyond any individual disease. If the personalised cancer vaccine paradigm matures as its proponents anticipate — and the early evidence is genuinely encouraging — it implies a future in which a significant fraction of cancer is treated not with broad-spectrum chemotherapy but with a bespoke immunological instruction set designed for each patient's specific tumour. If the HIV mosaic vaccine programme succeeds, it provides a template for attacking other highly mutable pathogens that have resisted conventional vaccine strategies. If mRNA delivery of gene editing machinery becomes routine, it accelerates the genetic medicine revolution by removing one of its principal remaining barriers.
The next 30 years will be determined by how well we build the second layer of that infrastructure: the manufacturing scale, the clinical trial networks, the AI tools, the regulatory frameworks for personalised medicine, and most importantly, the global health equity systems that determine whether this revolution belongs to everyone — or only to those who can afford it.
Buy me a coffee