Meet with us at ESGCT 2026

P0541 - Wednesday, October 28 , 1:30PM-3:30PM - Hall D
Synthetic Promoter Engineering Integrating AI-Guided In Silico Design, Biologically Driven Modeling, and Validation in CNS Systems, Dina Zielinski

P0131 - Wednesday, October 28 , 1:30PM-3:30PM - Foyer 3 + H
AI-Guided AAV Capsid Engineering and Multi-Modal Screening Pipeline for Targeted Podocyte Gene Therapy, Hassan Saei and Dylan Serillon

P0108 - Thursday, October 29 , 2:00PM-4:00PM - Foyer 3 + H
A Multimodal, End-to-End Rational Design Platform Breaks Native AAV Tropism Constraints, Diane Rebourcet

P0788 - Thursday, October 9 , 2:00PM-4:00PM - Foyer D to F
Accelerating Stable AAV Producer Cell Line Development Through AI-Driven Decision Intelligence, Yanis Habtoun

Authors: M. Morrier, J. Gatineau, P. Vidal, D. Zielinski

WhiteLab Genomics, FUTURE4CARE, 8 rue Jean Antoine de Baif, 75013 Paris, France

Getting a gene therapy to the right cell is only half the challenge. Once it arrives, the transgene still has to switch on in the right place, at the right level, and stay off everywhere else, a problem that becomes critical in the central nervous system, where densely packed, diverse cell types leave little room for error.

The promoter, the sequence that controls when and where a gene is expressed, is one of the most powerful levers for this kind of precision, yet it remains one of the most underused. Designing synthetic promoters from scratch is hard: rule-based methods miss the complex combinatorial logic of real biology, while purely data-driven models demand large datasets that simply don’t exist for most therapeutic targets.

With ALFRED, our AI-guided platform for genomic medicine design, we took a hybrid route, combining biological knowledge with data-informed modeling to design synthetic promoters even where data is scarce. Starting from our body-wide single-cell atlas and layers of epigenomic signal, the platform identified neuron-specific regulatory programs, assembled candidate promoters, and put them to the test in neuronal models.

The result: compact synthetic promoters that drive strong, selective expression in target neurons while staying quiet in off-target tissues, and a design-build-test-learn loop that gets sharper with every round.

Discover the full results at our poster at ESGCT 2026.

Authors: H. Saei1, D. Serillon2, C. Arrondel1, B. Estebe1, C. Gaborit1, A. Ammar1, C. Alliot2, B. Dafniet2, X. Liu2, A. Silvert2, P. Vidal2, G. Dorval1, G. Mollet1, C. Antignac1

1 Laboratory of hereditary kidney diseases, Inserm U1163, Institut Imagine, Université Paris Cité, Paris, France

2 WhiteLab Genomics, FUTURE4CARE, 8 rue Jean Antoine de Baif, 75013 Paris, France

Many severe kidney diseases still have no cure, and some of the most devastating trace back to a single, highly specialized cell: the podocyte. These cells form the kidney’s filtration barrier, and when they fail, a wide range of serious nephropathies can follow, including inherited forms of nephrotic syndrome. More than 60 genes expressed in podocytes have already been linked to monogenic kidney disease, making them a compelling target for genomic medicine.

The problem is getting there. AAV vectors are a safe, durable way to deliver genomic medicine, but today’s AAV serotypes simply don’t home to podocytes. Reaching these cells selectively has remained an unmet need.

The WIDGeT consortium set out to change that. In this poster, we focused on the work with the Imagine Institute, who brings expertise in rare kidney disease and representative organoid models. Our AI-guided receptor discovery platform identified a new podocyte-enriched receptor suited for AAV targeting. We then designed a 100,000-variant capsid library with AI that optimizes receptor binding and structural stability at the same time, so that more than 90% of designed variants could still be produced, even with challenging peptide insertions. Institut Imagine’s multi-modal renal screening pipeline then narrowed the field, surfacing over a thousand enriched variants and a first set of candidates showing selective binding and gene expression in cell models.

With organoid and in vivo studies now underway, this collaboration is building a path toward AAV vectors that can finally reach the kidney’s most elusive cells.

Discover the full story at our poster at ESGCT 2026.

Authors: D. Rebourcet1, B. Dafniet1, C. Alliot1, C. Colas1, E. Lazzarin1, J. Maes1, Y. Vander Meersche1, A. Barberis1, A. Silvert1, X. Liu1, P-L. Boudreault2, A. Murza2, F. Piguet3, J. Cottineau1, D. Serillon1, P. Vidal1

WhiteLab Genomics, FUTURE4CARE, 8 rue Jean Antoine de Baif, 75013 Paris, France

AAV vectors are a cornerstone of genomic medicine, but they come with a stubborn limitation: they naturally go to a handful of tissues, and getting them to target something new usually breaks them. Re-engineering a capsid to hit a new receptor tends to destabilize it, hurting both performance and the ability to manufacture it at scale. For a long time, retargeting and manufacturability have been a trade-off.

With ALFRED, our AI-guided platform for designing genomic medicines, we set out to remove that trade-off. The platform runs the whole chain in one workflow, finding the right receptor to target, designing the peptide that binds it from a library of over a million candidates, and engineering the capsid so it stays stable and easy to produce.

The results point to a real shift. Across targets in the kidney, brain, eye, and immune system, the platform reached hit rates far above what’s typically reported, with top binders hitting nanomolar affinity straight away. More than 90% of designed capsids were successfully produced, even with challenging insertions, and yields matched or beat the original vectors, so retargeting no longer comes at the cost of manufacturability. A prediction model now links design to real-world productivity with striking accuracy, and in vivo studies in mice confirm the vectors behave as designed.

The same design principles reach beyond AAV, toward other viral and non-viral delivery systems, including LNPs. It’s a step toward truly programmable vectors, designed to go exactly where they’re needed.

Discover the full results at our poster at ESGCT 2026.

Authors:C. Niehus1, C. Silva1, J. Drushku1, M. Hussong1, R. Whitfield1, Y. Habtoun2, N. Hirigoyen2, S. Berthoumieux2, D. Halie2 and J. Cottineau2

1 WhiteLab Genomics, FUTURE4CARE, 8 rue Jean Antoine de Baif, 75013 Paris, France
2 Cytiva, 26-28 Avenue de Winchester, 78100 Saint-Germain-en-Laye, France

Building a stable cell line that reliably produces AAV is one of the hardest, slowest steps in genomic medicine manufacturing. Success depends on a tangle of factors, culture conditions, selective pressure, and how cells adapt over time, and today most of that is worked out one experiment at a time. The result is long timelines and a lot of effort spent on cell lines that were never going to make it.

Together with Cytiva, we set out to bring data and AI to this problem. We built a framework that learns from more than 2,000 minipool generations and production runs, pulling together cell growth, viability, process data, operator steps, and final AAV yield into a single picture, then ranking what actually drives success at each stage.

The payoff is earlier, better decisions. Using only data from the first part of the process, the model can flag runs that are heading for failure, cutting wasted experiments by half, a result confirmed on an independent dataset. A second model predicts productivity and surfaces the process choices that matter most, turning cell line development from trial and error into interpretable, data-driven decision support.

Discover the full results at our poster at ESGCT 2026.

We'll be connecting with innovators throughout the event. Find us at the conference or book time with us bd@whitelabgx.com