WhiteLab Genomics Creates AI-Designed Viral Vectors with 50-Fold Increase in Gene Delivery to the Brain
At the annual conference of the American Society of Gene & Cell Therapy (ASGCT), data were presented on new adeno-associated virus (AAV) vectors designed by AI. Compared to AAV9, they achieve 50 times higher concentration of genetic material in the mouse brain with minimal liver uptake.
AI-DESIGNED VECTORS: When the Protein Is No Longer Natural
The Gist: What's Really Happening
At the annual conference of the American Society of Gene & Cell Therapy (ASGCT) in May 2026, the French company WhiteLab Genomics presented data that quietly but fundamentally change the game in gene therapy.
Their AI platform designed new variants of adeno-associated virus (AAV) capsids that deliver genetic material to the mouse brain 50 times more efficiently than the "gold standard" AAV9.
But the main point isn't the 50-fold improvement. It's the redistribution.
Traditional AAV9 reaches the brain only in minimal amounts, with the bulk of the dose accumulating in the liver. This creates two problems: first, huge doses must be administered (up to 1e14 viral particles per kilogram of body weight), which is expensive and risky. Second, the liver is the primary organ where immune responses against the capsid arise.
WhiteLab's new vectors solve both problems simultaneously. They not only deliver genes to the brain 50 times more efficiently, but also show significantly less liver uptake. Exact figures were not disclosed in the press release, but the phrase "minimum liver uptake" is the key innovation.
Timeline and Context
2019: WhiteLab Genomics is founded in France with support from Y-Combinator. Their idea: use machine learning to design genetic constructs, not just analyze data.
2021–2023: The company develops its CapsidMap™ platform—a system based on protein language models and reinforcement learning to design new AAV capsid variants.
2024: WhiteLab enters partnerships with Sanofi (ophthalmology and rare diseases) and Flagship Pioneering. They also work with Genethon and the Brain Institute in Paris—among the leading neurogenetics centers in Europe.
May 2025: At ASGCT 2025, the company first presents preclinical data on its AI-designed vectors. Results in mice show significant improvement in CNS delivery.
May 2026: At ASGCT 2026, new data follow—confirming 50-fold improvement and minimal liver uptake.
What remains behind the scenes: these vectors are not designed "from scratch" in a black box. WhiteLab's platform uses multiple approaches simultaneously: predicting receptor-ligand interactions for tissue specificity, transcriptomic analysis to identify receptors on target cell surfaces (e.g., specific neuron types or microglia), and generative AI to create new amino acid sequences for capsid proteins.
Who Wins and Who Loses
Winners (obvious):
- WhiteLab Genomics. After this presentation, the company becomes a leader in AI-driven capsid engineering. Their business model is partnerships with pharmaceutical companies, not developing their own drugs. Each successful partnership brings upfront payments, milestones, and royalties. Typical deals in this space: $5–20 million upfront, up to $200–300 million in milestones, and low-single-digit royalties.
- Patients with neurological diseases. Huntington's disease, spinal muscular atrophy (SMA), lysosomal storage disorders, some forms of Parkinson's—all could benefit from more effective and safer gene therapy. A 50-fold dose reduction means not only lower cost (AAV production is one of the most expensive steps) but also lower immunogenicity.
- Sanofi. As WhiteLab's partner in ophthalmology and rare diseases, they get exclusive access to these vectors for their target indications. For Sanofi, which is trying to regain ground in gene therapy after issues with their SMA program, this is a strategic asset.
Losers:
- Dyno Therapeutics. This competitor (a Harvard spin-out) uses a similar approach—AI for AAV capsid design. Dyno has raised over $100 million from investors including Andreessen Horowitz. WhiteLab's success at ASGCT means Dyno is no longer the only player in this niche, and pharma companies may start "shopping" among multiple suppliers.
- Manufacturers of non-viral vectors (lipid nanoparticles, LNPs). If AAV can be made so efficient that it reaches target tissues at ultra-low doses, the advantages of LNPs (lower immunogenicity, possibility of repeat dosing) become less critical. However, complete replacement won't happen—LNPs have their own niches (e.g., for RNA vaccines and in vivo gene editing).
- Research groups still using random mutagenesis to improve AAV. The traditional approach—creating a library of random capsid mutations and screening in animals—works but requires enormous resources (millions of dollars per library). AI design is radically cheaper and faster. Such groups will either have to adopt AI or remain in an academic niche.
What the Media Isn't Saying
Non-obvious insight #1: "50-fold" in mice does not equal "50-fold" in humans
No one mentions this in press releases, but it's a critical caveat.
Mouse AAV9 and human AAV9 are the same virus. But receptors on mouse and human cell surfaces may differ in affinity and distribution. The LY6A protein, which serves as the main receptor for AAV9 in mice, is expressed differently in humans.
If WhiteLab's new vectors were designed and optimized on mouse models (which is most likely), their efficacy in humans could be significantly lower. Or, conversely, higher—but the risk is unpredictable.
WhiteLab positions its platform as capable of accounting for human biology through transcriptomic analysis. But there is no data yet on whether these vectors have been tested on human tissues ex vivo or in non-human primates.
Non-obvious insight #2: Minimal liver uptake is not just about safety, but also manufacturing scalability
Here's what doesn't make headlines.
One of the biggest challenges in AAV therapy is manufacturing. The virus is produced in HEK293 or Sf9 cells, and the yield of the target product is often low. Some virions are "immature," some are empty capsids without a genome.
If the new vector is 50 times more efficient, it means 50 times fewer viral particles need to be produced to achieve a therapeutic effect. This radically reduces manufacturing costs—from the typical $500,000–1,000,000 per dose for some indications down to $10,000–20,000.
Moreover, fewer viral particles mean less burden on purification, lower risk of aggregation, and a more stable final product. This is not just a scientific breakthrough—it's a manufacturing breakthrough that makes gene therapy accessible to more patients.
Non-obvious insight #3: The company has not yet published a peer-reviewed paper with these results
A presentation at ASGCT is not a publication in Nature Biotechnology. It's an oral talk at a conference, which may be accompanied by posters but not full data.
WhiteLab was founded in 2019. In 7 years, the company has published very little in peer-reviewed journals, if at all. Their website does not have a "publications" section. This is normal for a startup focused on technology development rather than an academic career. But for the scientific community, it means the data have not yet undergone independent review.
Until the results appear in a journal like Molecular Therapy or Nature Communications, the "50x" figure should be taken with caution.
Forecast: Next 30 Days and 90 Days
30 days (by end of June 2026):
- Wave of interest from pharma. After the ASGCT presentation, WhiteLab will receive partnership inquiries from several large pharmaceutical companies working on neurogenetics. These could include Roche, Pfizer, Novartis (all have programs for SMA and other neurodegenerative diseases). Typical timeline for starting negotiations is 2–4 weeks after a major presentation.
- Data detailing. Expect the company to publish either a preprint or additional information on specific metrics in the coming weeks—exact comparison with AAV9 for neuronal transduction, distribution data in other tissues (heart, muscle, kidneys), and data on long-term persistence and immunogenicity.
- Competitor reactions. Dyno Therapeutics and other players (Codexis, 4D Molecular Therapeutics) will release communiqués about their achievements, trying to draw attention back to themselves. There will be a "press release war" over whose vectors are better. Without independent benchmarking, this remains marketing.
90 days (by end of August 2026):
- Start of non-human primate (NHP) studies. The next logical step is testing these vectors in non-human primates, whose CNS anatomy and receptor profile are closer to humans. Such studies typically last 3–6 months. WhiteLab may announce the start of such studies in partnership with a CRO (contract research organization) for $1–3 million.
- New funding round. WhiteLab Genomics will likely raise a new round—Series B or C. Previous rounds were not publicly disclosed, but the company was probably valued at $50–150 million. After ASGCT, the valuation could increase 2–3 times. Round size: $50–100 million from venture capital firms specializing in AI in biotech (e.g., Flagship Pioneering, Andreessen Horowitz, Lux Capital).
- Expansion of patent portfolio. The company will file new patent applications for specific amino acid sequences of the designed capsids. Patenting AAV capsids is a complex area, as many sequences may be deemed "obvious" based on known variants. But if WhiteLab's platform creates fundamentally new sequences with unpredictable properties, patents may be granted. Patenting costs in the US and Europe range from $50,000 to $200,000 per patent family.
Main forecast:
In 18–24 months, we will see the first clinical trials using these vectors.
Which of WhiteLab's partners will be first to enter the clinic? The most likely candidate is Sanofi, which already has gene therapy programs for ophthalmological and neurological diseases. Their experience with Zolgensma (AVXS-101) gives them the infrastructure to move quickly from preclinical to clinical stages.
If trials confirm safety and efficacy in humans, this will be the moment when an "AI-designed biological entity" first reaches the market in gene therapy. And that would be no less a breakthrough than the creation of gene editing itself. Because without effective delivery, CRISPR is just a beautiful tool without application.
WhiteLab Genomics does not develop drugs. They develop delivery methods for drugs. And if they are right, their delivery method will become the new standard for the entire industry.
The irony is that the company's founders are probably not biologists by training. They are computer scientists who learned to speak the language of proteins. And that is the real revolution—not a new molecule, but a new way of inventing molecules.
— Editorial Team
