On 12 August 2026, the David Liu lab at the Broad Institute published OptiPrime, a machine-learning model that predicts how well prime-editing guide RNAs (pegRNAs) will perform and ranks the strongest candidates before any are made in the lab (Broad Institute; Nature Biotechnology). It is the first prime-editing model to embed the biochemistry of the reaction into its structure, predicting the rate of each biochemical step and combining them into a single score.
In one demonstration, the team used OptiPrime to design a prime editor that corrected a pathogenic Kif1a mutation in a mouse model of a severe neurological disorder. It reached about 40% correction in the brain cortex, and optimization from the starting pegRNAs took roughly four weeks rather than months (Broad Institute). For teams developing genetically stratified therapies, the result narrows one long-standing constraint: the time and lab effort needed to design a working editor. It leaves a second constraint untouched, and that second constraint more often decides whether a precision program enrolls on schedule.
Prime editing, first described in 2019, can in principle correct the vast majority of known disease-causing variants. The estimate widely cited for the reachable share is about 89% of variants associated with human disease. The method has advanced on several fronts in 2026, including more stable pegRNAs, an AI-redesigned reverse transcriptase, and improved lipid-nanoparticle delivery.
The clinical picture is earlier than the science. Prime editing is not yet widely used in vivo, and the only publicly announced clinical application to date uses an ex vivo approach, editing cells outside the body before returning them. OptiPrime is positioned to support the newly launched Center for Therapeutic Genetics, which develops editors for rare and ultra-rare diseases, and it is free for non-commercial use. Importantly, gene-editing science tends to outrun its clinical translation, limited by cost, scalability, and unequal access to sequencing and variant data. OptiPrime is a clear example of the design side of that gap closing.
Design is one input to a genetic medicine. Enrollment is another, and it fails more often. Only about one in five trials find enough participants within the predefined timeframe, and adding a genetic eligibility requirement narrows the qualifying population further.
Each candidate therapy built with a tool like OptiPrime is defined by a specific variant or a small set of variants. The patients it can treat must be found, genetically confirmed, and consented before a single dose is given. Those steps run on infrastructure that AI-assisted design does not accelerate: outreach to dispersed and often undiagnosed populations, prescreening, and testing that returns a confirmed genotype. When genetic testing sits apart from recruitment, sponsors lose visibility into where candidates drop out, and eligible patients drop out at the handoffs between disconnected vendors.
As editor design gets faster and cheaper, the binding constraint on a precision program moves downstream, to identifying and qualifying the right patients. That is the workflow Sano connects, joining patient finding, prescreening, genetic testing, and engagement in one system rather than a patchwork of separate vendors.
The difference shows up in screening performance. In a multi-country genetic hearing-loss program across the US, UK, and Spain, more than 70% of referred candidates completed prescreening and about 90% of those met eligibility, because qualification and at-home genetic testing were part of the same participant journey rather than a separate operational step. When the design pipeline shortens to weeks, the programs that move fastest will be the ones that can confirm eligible patients at a matching pace.
Matching enrollment to a faster design cycle is an operational problem, not a scientific one. It rests on a few connected steps. Reaching the right patients means working across digital channels, patient communities, provider networks, advocacy groups, and genetic databases, because rare and genetically defined populations are rarely concentrated in one place. Qualifying them means prescreening that is genetics-aware, so candidates are assessed against the specific variant a program targets rather than a broad clinical definition.
Confirming them means genetic testing that returns a usable genotype without adding site or patient burden. At-home saliva collection lets geographically dispersed participants provide a sample in under ten minutes, with kit logistics, result tracking, and counselling handled inside the same workflow. Keeping testing, referral, and prescreening in one system gives sponsors traceability from first contact to confirmed eligibility, which is where fragmented programs lose both candidates and visibility. When each of these steps is measurable, a sponsor can see where a genetically defined cohort stalls and adjust before the delay reaches the site.
Two questions follow from OptiPrime. The first is whether patient-side infrastructure can scale as quickly as design tools now can. A four-week editor is only useful once there are confirmed patients to treat, and confirmation depends on sequencing, referral networks, and testing logistics that no design model touches.
The second is how ultra-rare programs handle the diagnostic gap. Many eligible patients are undiagnosed, or carry variants that no one has yet linked to disease, so broader access to sequencing and variant data will shape which of the reachable variants translate into a treatable patient population. OptiPrime lowers the cost of asking the design question. The answers still depend on finding the people the answer is for.
Get in touch to discuss patient identification and genetic testing for a genetically stratified program.