How Prime Medicine scientists used Benchling’s AI Scientist to accelerate PM359’s path to BLA

Prime Medicine will be speaking about their collaboration with Benchling at the 2026 ASGCT Annual Meeting, Monday May 11th, as part of the Automation in CGT Manufacturing workshop.

We recently laid out a vision for an AI Scientist “that deserves the name”: one that operates across the entire design-build-test-learn loop, connects the digital and physical worlds, and reasons over an organization’s experimental history.

AI Scientists are already beginning to reshape discovery workflows, from target identification to candidate optimization. But some of the most challenging bottlenecks emerge during development. Our collaboration with Prime Medicine shows what an AI Scientist brings to one of the hardest phases of taking an advanced modality to market: the final CMC push toward a BLA.

When clinical momentum outpaces CMC

Prime Medicine is advancing PM359, a prime-edited CD34+ HSC therapy for chronic granulomatous disease (CGD), toward a BLA. CGD is an ultra-rare, life-threatening immunodeficiency, and PM359 has generated exceptional clinical data. Early results showed rapid engraftment and restoration of immune function to levels well above the threshold believed to be potentially curative.

These results have created an urgent obligation. CMC must accelerate to match the clinical momentum. But PM359 is far from a conventional biologic.

PM359 consists of three RNA components delivered into autologous CD34+ hematopoietic stem cells, each with its own long chain of critical starting materials. To support a BLA, Prime must also validate 35 product-specific assays to FDA standards. That is a formidable analytical validation effort under any timeline, let alone an accelerated ultra-rare one.

The traditional path, and why it doesn’t fit

Analytical method validation is traditionally prospective: define every validation parameter up front, then execute dedicated studies from scratch. That effectively sets aside the method development and qualification work done to support the IND. Validation starts fresh, by design.

For a standard commercial program with abundant manufacturing runs and longer timelines, that approach is workable. For PM359, it’s prohibitively slow.

The FDA’s May 2026 guidance for cell and gene therapies recognizes this tension. It supports leveraging historical data and manufacturing experience to reduce redundant experimentation. The challenge is proving, rigorously and traceably, which requirements are already met by existing evidence and then guiding targeted follow-up studies. That’s where an AI Scientist can help a scientist move faster.

Prime’s novel approach, supported by the AI Scientist

Rather than setting aside years of accumulated work, Prime treated that evidence as the validation dataset, using the AI Scientist to synthesize and analyze it, design follow-up studies, and draft a validation report ready for review.

This was possible because Prime had been capturing every experiment in Benchling since 2022: analytical test results, process development studies, manufacturing of GLP test article materials, assay development and qualification, and CDMO tech transfer reports.  

For the cell-based potency assay, AI Scientist harvested data across 50+ experiments in their electronic lab notebook, with every data point traceable to its source. It then mapped each data point to the relevant ICH Q2(R2) validation performance characteristic and Prime's internal guidance, determining which requirements the existing evidence already satisfied and where additional studies were still needed.

For complex characteristics like intermediate precision, it fit a mixed-effects variance components model to estimate the long-term assay variability from historical production data. For robustness, it analyzed historical DOE data to identify opportunities to tighten operational ranges, and expand differentiation windows, then designed targeted follow-up studies to validate each recommendation.

Closing the loop from analysis to bench

Identifying validation gaps is only half the problem. The follow-up studies still have to be executed in the lab.

For each gap, the AI Scientist generated the experimental templates: protocol attached, acceptance criteria pre-specified, ready for automated lab execution. As studies completed, it ingested the results and updated the validation data dynamically, producing a living package ready to be reviewed and traceable to every source.

What would traditionally require months of manual curation and analysis was compressed into days, with the AI Scientist amplifying rather than replacing the scientific judgement that only Prime’s team could provide.  That is the future the FDA’s guidance was written to enable, made possible by structured scientific data.

The foundation that made it work

Prime's early investments established a model that scales: developing a functional potency assay, leveraging lab automation for precision and reproducibility, and capturing that work in structured data. As Prime advances into new prime editing therapies, each requiring its own potency assay to reflect a distinct mechanism of action, that foundation means the AI Scientist is going beyond accelerating validation and becoming a partner in assay development itself.

As AI Scientists become more capable, the organizations with rich scientific context, strong data foundations, and a willingness to embrace AI will move the fastest. Prime Medicine is one of them.

To learn more about the AI Scientist, visit benchling.com/ai or join the waitlist.