AI & Protein Engineering

September 22, 2026 by

The biggest drivers of cell therapy cost of goods (COGs) are decisions made during process development, not manufacturing execution. Media consumption, manual touchpoints, facility throughput, and analytical testing burden are all downstream consequences of upstream design choices. Reducing COGS at scale without introducing quality risk means getting those early decisions right, not optimizing around them later.

The Process Design Decisions That Lock In Cost

Programs do not set out to build expensive processes. They build processes optimized for speed at discovery stage with high media volumes, manual interventions, custom analytic panels, and then carry those designs forward into clinical manufacturing. By the time anyone models the cost per dose, the process is fixed.

The decisions with the largest COGS impact are often made in the first months of process development: choice of activation method, media formulation and feeding strategy, expansion platform, and harvest timing. Each of these determines how much labor, material, and cleanroom time a single batch requires. A process designed around manual sampling and open manipulations does not become closed and automated as the program advances. It becomes expensive.

Labor and Facility Time Are the Levers Most Programs Underestimate

When reviewing program costs, raw materials get the attention, but it is the number of manual steps per batch and the days each batch occupies a suite that determine whether a process can scale economically. In autologous cell therapy, labor and facility occupancy typically account for 50–70% of COGS. A process that requires twelve operator interventions over nine days in a cleanroom has a fundamentally different cost structure than one designed for six interventions over six days, even if the cell product is equivalent.

Programs that address this early do three things: design for closed processing, reduce in-process sampling, and select expansion platforms that shorten culture duration and/or employ automation. That discipline builds cost efficiency into the process rather than trying to engineer it in after clinical data is generated.

ElevateBio Is Building Toward Manufacturing Efficiency as a Design Principle

At ElevateBio, the investment in manufacturing technology and platform architecture is oriented around this problem. Process development, MSAT and manufacturing operate as a connected system, so that decisions about cost, throughput, and quality are made together rather than sequentially. The goal is not to optimize a process after it is built. It is to design processes where efficiency is structural — where closed systems, shortened culture durations, and streamlined analytics are built in from the start, not retrofitted under commercial pressure.

The Programs That Reduce COGS Are the Ones That Design for It

Cost reduction in cell therapy is not a manufacturing problem. It is a process development problem. The programs that reach commercial scale with viable unit economics are the ones that treated COGS as a design constraint from day one, not the ones that built a process first and asked what it costs later. The earlier manufacturing economics enter the conversation, the more room there is to act on them.

Chris Shumway, M.S.
Senior Director, Manufacturing

Chris Shumway joined ElevateBio in 2024 as Senior Director of Manufacturing, where he oversees manufacturing processes at ElevateBio BaseCamp, ensuring efficiency, compliance, and high-quality production. He has more than 20 years of experience in process development, validation, quality systems, and driving continuous improvement in manufacturing operations.

Prior to his role at ElevateBio, Chris held multiple manufacturing and leadership positions at Avid Bioservices, NantKwest, and AbbVie. He received his M.S. in Engineering Management from Syracuse University and his B.S. in Chemical Engineering from Lafayette College.

September 16, 2026 by

One assay is not enough for gene editing off-target safety assessment because each method detects a different subset of potential edits, and off-target profiles can change by editing modality, cell type, donor context, and dose. The most comprehensive safety profiles combine biochemical, computational, and cell-based approaches, followed by deep sequencing confirmation and biological risk annotation.

Different Methods Detect Different Off-Target Events

Biochemical approaches such as Digenome-Seq help nominate potential off-target sites by testing editor activity against genomic DNA. Computational tools identify sites based on homology patterns. While both assays are useful for nominating genomic sites for further safety evaluation in a cell-free manner, a direct measure of in-cell editing activity is missing. In our assessments at ElevateBio, whole exome sequencing (WES) identifies off-target events down to 1% allele frequency. WES acts as an unbiased safety net, nominating and confirming edits across all coding regions without requiring prior knowledge of where to look.

Off-Target Profiles Can Change by Cell Type and Editing Modality

The same guide RNA and CRISPR editor can generate different off-target patterns in different cell types, for instance primary T cells versus primary hepatocytes. A nuclease and base editor using identical guides can create distinct off-target signatures. These biological variables mean that safety assessment in a single cell type or with a single detection method provides false confidence. Primary cells from both male and female donors reveal the true editing landscape that immortalized cell lines may mask.

Assessment Data Should Feed Back Into Editor Engineering

Off-target reduction happens through iterative cycles of detection and protein engineering. At ElevateBio, improvements to adenosine deaminase domains have reduced both the number of off-target sites and editing magnitude at each site. This optimization is enabled by a constant feedback loop between off-target assessment and engineering to accelerate the development of safer, more specific therapeutics.

Orthogonal Methods Turn Detection Data Into Safety Insight

Combining Digenome-Seq, computational prediction through tools like Calitas, and whole exome sequencing captures the broadest spectrum of potential off-target events. Each confirmed site undergoes clinical variant analysis using ACMG/AMP guidelines and comparison to databases of known genetic variation to determine biological risk (see the technical poster, with data). Dose-response curves in primary cells establish whether off-target editing occurs at therapeutically relevant doses. This integrated framework transforms raw detection data into risk assessments that guide both engineering decisions and regulatory submissions.

Off-target safety assessment should not rely on one assay or one model system. Programs that generate the most useful safety profiles layer orthogonal detection methods, evaluate therapeutically relevant primary cells, and use the results to guide editor engineering before IND-enabling decisions are locked.

Allie Crawley, PhD, Associate Director, Computational Biology

Allie oversees teams building applied genomics platforms in the CRISPR medicines field. She has worked to establish bioinformatic software and workflows, including design of correction strategies for genetic diseases, evaluation of editing, and off-target safety measurements for genomic medicines. After receiving her PhD in Functional Genomics from North Carolina State University, she joined AgBiome and helped build the foundational technology platform owned by ElevateBio Life Edit.

September 16, 2026 by

Potent, tolerable, and development-ready gene editing is not achieved through a single design decision. It improves through iteration: optimizing the editor, RNA chemistry, guide design, delivery vehicle, and target biology as one connected system.

This is especially true in the liver. LNP-mediated delivery has created a practical path for transient RNA delivery, but performance still depends on many interacting variables. A strong editor is not enough if the RNA composition is weak. A potent lipid nanoparticle (LNP) is not enough if the guide is underperforming. Development depends on making each component better, then carrying those learnings forward.

Drug Substance Optimization Can Unlock Potency

For in vivo editing, the drug substance is more than a single variable. It is the combined performance of the RNA payload, guide RNA, chemical modifications, and their compatibility with the LNP delivery system.

At ElevateBio, HAO1 has served as a useful in vivo liver target for optimizing LNP-RNA drug substances. In mouse studies, internal mRNA modifications and guide RNA modifications increased editing potency, with individual changes producing meaningful gains. These results show why optimization cannot stop at editor selection (see presentation and data). RNA design and chemistry can materially change in vivo editing outcomes.

Cross-Modality Learning Accelerates Development

The value of an integrated platform is that improvements in one modality can inform the next. Learnings from HAO1-targeted nuclease experiments helped guide more complex RNA payload strategies, including RT editing. In mouse studies, our internal formulation, LNP1, delivered an RT editor targeting HAO1, and the addition of an accessory protein improved editing.

The same feedforward logic supported base editing development. A proprietary LNP and adenine base editor were used to target PCSK9 in vivo, producing efficient editing with reductions in serum PCSK9 and total cholesterol. Those learnings helped inform subsequent NHP-ready editing work.

In Vivo Readiness Requires More Than Potency

Potency matters, but it is only one part of the development equation. Advanced therapy programs also need tolerability, translatability, and a path to improved follow-on editors.

ElevateBio’s LNP1-delivered A-base editor produced functional editing and target protein reduction in NHPs, with minimal transient liver enzyme changes across evaluated doses. Platform feedforward then enabled faster selection of follow-up editors, including engineered variants and alternate systems that improved in vitro potency.

In vivo gene editing advances fastest when optimization is iterative and integrated. By connecting RNA design, guide chemistry, LNP formulation, protein engineering, and preclinical validation, ElevateBio helps partners move from early editing concepts toward in vivo-ready programs with greater speed and higher potency without compromise.

Tyler McCullough, Ph.D

Tyler McCullough, Ph.D., is the Associate Director of Translational Biology at ElevateBio’s Life Edit Therapeutics. He brings 11 years of extensive experience in gene editing with a focus on Lipid Nanoparticle (LNP) and Adeno-Associated Virus (AAV) delivery systems targeting various organs, including the eye, muscle and liver. His expertise includes IND-enabling studies, from mice up to non-human primates (NHP).

Previously, Dr. McCullough held several roles at Life Edit, such as Senior Scientist, Preclinical Development. In this role, he was responsible for conducting preclinical studies, including first-in-mouse editing with proprietary systems and delivery methods. His work helped generate critical in vivo editing data that contributed to securing key partnerships.

During his academic career, Dr. McCullough collaborated with industry leaders in gene editing, independently designing and executing studies that led to one of the first therapeutic CRISPR-based editing studies in NHPs. He completed his post-doctoral fellowship at Duke University and earned his Ph.D. in Genetics from the University of Florida. Additionally, he holds a Master of Health Science from John Hopkins University and Bachelor of Science in Microbiology and Cell Science from the University of Florida.

August 11, 2026 by

AI-directed protein engineering can optimize proteins for genomic medicines when machine learning models are trained on high-quality experimental activity data and paired with rapid design-build-test cycles.

At ElevateBio, we combine automation, protein language models, and active learning to identify protein variants with improved activity for genomic medicine applications.

Gene insertion by large serine recombinases (LSRs) or retrotransposases can make safer, more potent CAR T cell therapies and can be the solution to broadly curing monogenic diseases caused by a variety of DNA changes within the patient population.

Better Starting Proteins Create a Stronger Engineering Foundation

Our computational biologists mined ElevateBio’s large, curated bioinformatic databases to identify a diverse set of novel large serine recombinases and their corresponding attachment sites. Most of these wildtype proteins mediated large gene insertion in human cells, with several outperforming the literature standard Bxb1 molecule. More protein diversity means we can select the optimal LSRs to engineer for each therapeutic challenge.

Engineering in Mammalian Cells Improves Human-Relevant Performance

Engineering in mammalian cells avoids the common trap of finding proteins that work well in bacteria, but not in humans. At ElevateBio, we use experimentally derived activity data from mammalian cells to intelligently discover more potent molecules than those selected by nature, which translate into superior therapies.

Protein Language Models Focus the Search on Higher-Value Design Space

The limited throughput of mammalian cell wet lab experiments makes probing the vast protein design space difficult. Our use of protein language models combined with experimental data enables exploration of the vast and otherwise unmanageable protein fitness activity landscape. By identifying high fitness protein domains near activity maxima, we can rapidly focus our search on protein modifications that improve function.

Automation Accelerates the Design-Build-Test Cycle

With ElevateBio’s automation platform, we can synthesize protein variants in-house, reducing cost and speeding up the design-build-test cycle. After two rounds of machine-learning guided protein optimization completed in just 4 calendar weeks, we found many single mutations that improved large gene insertion, including one that was 2.4-fold better than wildtype protein from nature. By generating pairwise combinations of the most beneficial mutations, we found a new double-mutant variant 3.4-fold more active than wildtype (view and download poster on this topic). These data additionally measured interactions between pairs of individual mutations to facilitate design of higher order mutants which will further improve activity.

The takeaway is clear: protein engineering is most powerful when combining protein language models, machine learning, automation, and mammalian cell-based activity profiling as one integrated system. At ElevateBio, that integrated system helps accelerate protein optimization for genomic medicine development and deliver more effective solutions for our partners.

August 11, 2026 by

Targeted large gene insertion is becoming one of the most important frontiers in genomic medicine. Many therapeutic applications benefit from or require inserting a full gene into a defined genomic location.

In many oncology or autoimmune therapies, this technology enables delivery of CARs. It also matters in monogenic disease, where inserting a functional gene may address multiple disease-causing variants with one therapeutic strategy.

The Field Needs Fit-for-Purpose Insertion Systems

Large gene insertion is not one problem. Payload size, delivery, genomic target access, insertion efficiency, specificity, and fit for either in vivo or ex vivo use are important considerations for the correct approach. No single enzyme class can address all potential applications.

Large serine recombinases (LSRs), CRISPR-LSR systems, and retrotransposons each bring different strengths. LSRs can insert large cargo without exposed double-strand breaks and are not limited by payload size. Retrotransposons offer an all-RNA delivery path. CRISPR-LSR approaches can make insertion more programmable, but with added complexity that may be better suited for ex vivo engineering.

The right question is not which modality is best. It is which modality best fits the biology, delivery route, target site, and development path.

Enzyme Diversity Creates More Paths to Success

Large insertion becomes more powerful when developers are not limited to a small set of naturally known enzymes. ElevateBio has built a discovery and engineering pipeline designed to expand that starting set.

Using a catalog of more than 10 billion proteins, ElevateBio mined more than 30,000 LSR candidates, and identified more than 100 active LSRs. Many demonstrated higher activity than the literature benchmark Bxb1 (view presentation and data).

This breadth creates optionality. Different LSRs have different recognition sites, activity levels, and integration profiles. A larger enzyme set gives developers more chances to find an insertion system that fits the therapeutic program instead of forcing the program to fit the tool.

Programmability Is the Next Barrier

LSRs are limited by the sites where they naturally integrate their cargo. Most natural or un-engineered LSRs do not have the desired insertion profiles into human genomes needed for therapeutic applications, which means the LSR technology cannot become broadly useful until engineered. ElevateBio is applying active learning, rational design, directed evolution, and generative AI to improve enzyme performance. The long-term goal is straightforward: choose the desired insertion location, then engineer or design an enzyme system matched to that site.

By leveraging our proprietary CRISPR technology to install landing pads, ElevateBio can make LSR-mediated large insertion programmable at defined genomic sites. We developed this CRISPR LSR approach in primary T cells and have shown installation of LSR landing pad with greater than 90% efficiency at the TRAC locus. Addition of our LSR components achieved 67% installation of a CD19 CAR under the endogenous TRAC promoter. This approach is an attractive avenue for ex vivo manufacturing applications that require superb programmability.

The Future Is Fit-for-Purpose Insertion

A potent and diverse targeted insertion platform gives developers multiple paths to the same goal: inserting the right cargo, at the right site, in the right cell type, using a system matched to the therapeutic need.

Large gene insertion is not just another editing modality. It is a way to expand what genomic medicines can be designed to do.

David Wiley, Director, Nucleic Acid Technology

David Wiley is the Director of Nucleic Acid Technology at ElevateBio. His work focuses on guide RNA engineering, mRNA engineering, and production optimization, and he is currently leading the development of next-generation large gene insertion technologies, focusing on R2 retrotransposons and large serine recombinases technology.

Before ElevateBio, David was at Prime Medicine, where he led efforts in pooled screening platform development and early mRNA optimization. Prior to that, he was at Vertex Pharmaceuticals, working in the Functional Genomics group to identify potential therapeutic targets using CRISPR-based pooled screening approaches.

David earned his B.S. in Cell Biology and Genetics from the University of Georgia and his Ph.D. in Cell and Molecular and Developmental Biology from the University of North Carolina at Chapel Hill. He completed his postdoctoral training at Boston Children’s Hospital.

In his free time, David enjoys carpentry and spending time outdoors.

July 21, 2026 by

Base editing will not reach its full therapeutic potential if the field is limited to enzymes nature has already evolved. To expand what adenine base editors can do, we need to design deaminases for the job, not just search for better natural starting points.

The Bottleneck Is No Longer Imagination, It Is Experimental Throughput

Protein sequence space is too large to explore exhaustively. Even ambitious directed evolution campaigns test only a fraction of what is theoretically possible, and many enzymes that show early activity still carry unwanted attributes, including bystander editing profiles that limit targetability. Better enzymes exist. The challenge is finding them quickly enough, and with enough precision, to matter.

Generative AI Breaks Evolutionary Constraints

We wanted to explore beyond evolution. We trained a generative model on 370,000 putative adenine deaminases curated from ElevateBio’s 10+ billion protein catalog. Every training sequence was validated computationally, including protein folding with ESMFold, then filtered for quality before inclusion in the training body. The model learned the design grammar of this enzyme family, then generated diverse sequences while maintaining the active site, binding sites, and core folds required for functional deaminases.

While generating new sequences, selection of sampling parameters is what separates useful diversity from noise. More deterministic settings produced high duplication rates and sequences nearly identical to training data. Balancing creativity with determinism produced more diverse protein sequences ready for experimental validation.

Generation, Filtering, and Optimization Need to Work as One Loop

Generating novel sequences is only the first step. We pair sequence generation with strict computational filtering and a machine learning optimization pipeline that uses experimental feedback to improve candidates quickly. Using the “generate, test, optimize” feedback loop, we mitigate much of the inherent risk of de novo generation by treating protein design as an iterative engineering problem.

Although here we chose to focus on base editors, this integrated platform enables us to design purpose-built editors that can be matched to specific therapeutic targets or disease mechanisms across the full spectrum of gene editing modalities.

One Round of Optimization Creates a Precision Editor

Our top de novo deaminase candidate started at less than 2% on-target A→G editing. After a single round of ML-based optimization, it exceeded 20%. The results:

Greater than 20% on-target A→G base editing after one optimization cycle (view poster and data)

Minimal bystander profile across a panel of therapeutically-relevant genes. Examples include less than 1% bystander editing just two nucleotides from the target base

The specificity profile is key. Bystander editing at nearby positions is the factor that makes many genomic targets intractable with current base editing tools. A deaminase that edits its target base while leaving the +2 position unchanged opens sites that were previously untouchable.

The next leap in base editing will not come only from finding better natural enzymes. It will come from designing fit-for-purpose deaminases, testing them quickly, and improving them through experimental feedback. That shift moves enzyme discovery from a search problem to an engineering discipline.

Matt Nethery, Principal Data Scientist, Computational Biology

Matt Nethery works as a Principal Data Scientist in ElevateBio’s gene editing services, responsible for the implementation of generative AI for discovery and engineering of novel genome editing tools. He has 15 years of wide-ranging industry experience, from building software applications and mobile analytics in the healthcare industry, to bioinformatics and generative AI. Matt completed his PhD in Functional Genomics at North Carolina State University researching microbial genomics, CRISPR biology, and phage engineering.

July 21, 2026 by

A base editor becomes potent and specific enough for therapeutic SNP correction when the right starting editor is selected from a diverse library, then optimized through parallel engineering of both the editor protein and its guide RNA. For therapeutic development, “close enough” is not enough. The editor must make the intended change efficiently while minimizing bystander and off-target activity.

The Right Starting Editor Creates a Stronger Foundation

You cannot engineer a system to its full potential if you start with a suboptimal tool. Instead of trying to force a single, one-size-fits-all editor to work for every target, the most effective approach begins with screening a diverse library of systems.

By comparing the activity of multiple editor families against a specific pathogenic target, you can identify the one with the most favorable inherent properties. ElevateBio uses this data-driven selection to provide a superior foundation for subsequent engineering.

Parallel Protein Engineering Drives Both Potency and Specificity

Once the best foundational editor is selected, the real work of optimization begins. At ElevateBio, our gene editing platform focuses on engineering the editor’s core protein machinery to boost its performance.

By creating and testing variants, we can significantly increase on-target editing efficiency while simultaneously designing out unwanted activity. This includes minimizing “bystander” edits at adjacent bases and off-target edits elsewhere in the genome, ensuring the editor makes the single, precise change required to address a partner’s disease target.

Guide RNA Optimization Unlocks Additional Performance

The editor protein doesn’t work alone. Its guide RNA (gRNA) determines where the editor goes and can strongly influence how efficiently it performs.

Guide RNA optimization provides a complementary path to improve potency. By optimizing the guide backbone sequence, length, and introducing modified bases, ElevateBio further improved editing performance (see the poster with data). In this program, the final optimized guide RNA was 2.5-fold more potent than the initial version, contributing to an overall improvement from ~10% to over 85% correction when combined with the engineered adenine base editor.

Therapeutic SNP Correction Requires Building the Editor Around the Target

Potent and specific base editing is not achieved by choosing an editor off the shelf. At ElevateBio base editors for therapeutic SNP correction are built around the biology, sequence context, and performance requirements of our partner’s specific disease target mutation.

Sean Crosson, Senior Scientist, Translational Biology

As a Senior Scientist at ElevateBio, Sean Crosson engineers and optimizes next-generation gene editing technologies, including base and reverse transcriptase (RT) editors, to develop therapeutics for rare diseases. His background in gene therapy also includes developing novel AAV capsids for ocular diseases at Atsena Therapeutics and during his postdoctoral fellowship.

Sean holds a Ph.D. in Biomedical Science from the University of Florida and dual B.S. degrees in Biology and Chemistry from UNC Chapel Hill. He spends his free time baking, woodworking, and with his family.

June 24, 2026 by

The Limits of Nature in Gene Editing

Gene editing has moved well beyond proof of concept, now with several approved medicines and the technology expanding beyond the original CRISPR-Cas9 approach. Multiple modalities of gene editing tools now exist, each designed to address different classes of genetic disease. Yet these tools face a common limitation: they were built from proteins shaped by evolution, not engineered for human therapeutics.

Consider base editing. Enzymes called deaminases enable precise single-nucleotide changes, but naturally occurring variants are constrained by off-target activity and unwanted “bystander” edits at neighboring positions. These limitations are the result of proteins that evolved for biological functions unrelated to human therapeutics. Expanding beyond what nature provides requires a fundamentally different approach to protein design.

ElevateBio’s Generative AI Platform: From Discovery to Design

Generative AI is transforming protein engineering by moving the field from adapting nature to designing it. ElevateBio’s gene editing platform demonstrates this shift, not only by expanding its natural protein library, but generating entirely novel proteins – and then optimizing them for therapeutic relevance.

The foundation is ElevateBio’s proprietary catalog of more than 10 billion proteins extracted from extensive genome and metagenome collections. This dataset provides the necessary breadth to train generative protein language models (PLMs) that produce proteins beyond what natural evolution has explored. Once generated, AI can also help guide the development of these proteins into effective gene editors, establishing an end-to-end workflow for designing purpose-built therapeutic proteins.

Proof of Concept: AI-Designed Deaminases for Base Editing

ElevateBio demonstrated this approach by designing novel adenine deaminases for base editing. Using ElevateBio’s generative AI trained on more than 370,000 curated protein sequences from its 10+ billion protein catalog, the platform generated de novo proteins with low sequence identity to known deaminase. These were genuinely novel sequences not found in nature.

After a single round of machine-learning optimization, the top candidate showed a 20-fold improvement in on-target A-to-G editing efficiency. Critically, this potency came without sacrificing precision: the optimized editor maintained less than 1% bystander effects at the critical +2 position, just two nucleotides from the on-target site.

These results were validated across a panel of therapeutically relevant gene targets, demonstrating that the minimal bystander profile is a generalizable property of this AI-designed protein’s architecture. Importantly, this optimized editor enabled targeting of previously intractable genomic sites, highlighting how AI can expand not only performance, but the range of diseases that may ultimately be addressable through base editing.

Beyond Deaminases: AI Applicability Across Editing Modalities

The implications extend well beyond a single class of enzymes. ElevateBio is already applying the same generative AI and machine learning capabilities to design novel large serine recombinases (LSRs) for targeted gene insertion addressing diseases that require insertion of full-length genes rather than single-base corrections.

Active learning-guided optimization has already demonstrated 3.4-fold potency improvements on recombinase proteins in mammalian cells, validating that the same ML pipeline proven on deaminases translates to entirely different enzyme classes and editing mechanisms.

As ElevateBio’s datasets grow and its models improve, this AI-enabled optimization workflow will be applied across its full spectrum modalities – nucleases, base editors, RT editors, epigenetic modifiers, and targeted insertion systems – with properties designed for specific diseases and delivery methods. By beginning with AI-enabled diversity rather than relying solely on naturally occurring biology, ElevateBio can identify editing systems optimized for distinct disease mechanisms and patient needs, while also accelerating development timelines for partners.

Redefining the Boundaries of Therapeutic Development

Generative AI is redefining the boundaries of protein engineering and therapeutic design. By enabling the creation of proteins tailored for specific diseases, delivery strategies, and clinical constraints, AI-driven engineering shortens the path from concept to clinic and expands the range of treatable diseases.

The future of gene editing is no longer limited by what evolution has already produced, but by what biology can now be intentionally designed to achieve.

Amy Pooler, Ph.D., Chief Scientific Officer

Amy is the Chief Scientific Officer of ElevateBio. She has more than 20 years of scientific leadership and expertise with an extensive background in genetic medicine, neuroscience, drug development, and scientific strategy. Prior to joining ElevateBio, Amy held various leadership roles at Sangamo Therapeutics. She most recently served as Vice President, Head of Research where she helped establish multiple global licensing and R&D collaborations. She also built the neuroscience team and established an internal neurology pipeline, in addition to supporting multiple programs partnered with pharma companies. Amy received her Ph.D. in Cellular and Molecular Neuroscience from the Massachusetts Institute of Technology and her B.S. in Neuroscience from Brown University.

Matt Nethery, Ph.D., Principal Data Scientist, Computational Biology

Matt Nethery is a Principal Data Scientist in ElevateBio’s gene editing services, responsible for the implementation of generative AI for discovery and engineering of novel genome editing tools. He has 15 years of wide-ranging industry experience, from building software applications and mobile analytics in the healthcare industry, to bioinformatics and generative AI. Matt completed his Ph.D. in Functional Genomics at North Carolina State University researching microbial genomics, CRISPR biology, and phage engineering.

September 9, 2024 by

The recent convergence of artificial intelligence (AI) and biology has spurred remarkable scientific advances, most notably AlphaFold’s ability to predict 3D structures of proteins from amino acid sequences. These advances have been fueled by the availability of data for training AI models and technology capable of processing data faster than ever before, and they demonstrate the power of AI to revolutionize biological research.

Gene editing technologies, especially CRISPR, are likewise revolutionizing the therapeutic landscape with the promise of curing diseases in a single dose by correcting their underlying genetic drivers. Given the myriad diseases that CRISPR could treat, AI is becoming indispensable for delivering on the promise of gene editing.

Life Edit, ElevateBio’s gene editing business, is uniquely positioned at the interface of computational science and CRISPR biology to do just that. Here, we examine why AI is essential to the future of CRISPR discovery, what challenges AI faces in the CRISPR development, and how Life Edit is addressing those challenges to accelerate CRISPR discovery and enable our partners to develop novel and life-changing genetic medicines.

AI is indispensable for future CRISPR gene therapies

The first wave of CRISPR therapies have focused on a small collection of monogenic diseases, such as sickle cell disease, where the underlying cause is a single SNP in a single gene that can be corrected with ex vivo editing of patients’ cells. Developing CRISPR systems to treat these diseases, without the aid of AI, is relatively straightforward with current experimental approaches. The real challenge lies in expanding CRISPR’s application to the thousands of more complex monogenic disorders, as well as the multitude of polygenic diseases involving multiple gene corrections. Treating all these diseases requires development of many custom CRISPR systems, which would take an extremely long time to achieve with conventional approaches.

A key barrier with conventional approaches is that – for reasons researchers do not fully understand – different CRISPR systems are often needed to edit different disease-associated loci; successful gene editing for each disease thus involves the lengthy process of engineering and optimizing both the nuclease and a guide RNA (gRNA) of a given CRISPR system. There is no quick and rational way to effectively predict and design a CRISPR system with the optimal editing window, potency and specificity for a disease – despite the large quantities of experimental data available in the literature and proprietary databases.

These databases are insufficient for the task because the spaces of possible CRISPR protein sequences (based on 20 amino acid letters) and possible RNA sequences (based on four nucleotide letters) are astronomically huge and cannot be fully explored by experimental means alone. We believe AI is well positioned to tackle this problem using large language models (LLMs). Our belief is encouraged by recent AI advances in small molecule drug discovery, where AI models, trained on high throughput screening data, can perform virtual screening to rapidly narrow the number of hits for experimental validation.

Training AI in the language of therapeutics

While LLMs such as ChatGPT represent a major breakthrough in AI’s ability to use everyday human language, AI models for CRISPR nucleases are not quite as developed. Current AI models trained on natural CRISPR sequences from public databases are capable of predicting CRISPR structures and even generating novel systems. However, upon experimental validation, many do not have sufficient editing activity in mammalian cells, and those that do have sufficient activity sometimes perform at a lower level of efficiency than the current benchmark nuclease, SpyCas9. We believe the gap between function and therapeutic utility might exist because bacteria evolved CRISPR to detect and destroy infecting viruses, not edit and correct human genes, so there is no evolutionary pressure to produce CRISPR systems that are efficient in mammalian cells for treating diseases.

To achieve AI models that have therapeutic utility, we need to train the models with a large amount data from wet-lab experiments that are specifically designed for improving editing activity and specificity in disease-relevant mammalian cells. These data should include sufficient information of systematic search in the CRISPR protein and RNA sequence space that can address potential bottlenecks in gene editing inside mammalian cells, which are probably multifaceted and cell- or disease-dependent. AI models can help us find the right systems for the right targets quickly, without having to understand all of the bottlenecks, and these models should be continuously improved by a feedback loop, in which model predictions are constantly evaluated by experiments. Only with therapeutically relevant AI models for CRISPR we will be able to adequately address the challenges of developing CRISPR systems for the vast majority of monogenic and polygenic diseases.

Life Edit’s solutions for bridging the CRISPR gap

Life Edit is an integral part of ElevateBio’s genetic medicine ecosystem of enabling technologies and end-to-end manufacturing capabilities. We’re also distinct in how we work compared to other gene editing companies. We started with the “top-down” approach of building a diverse array of CRISPR nucleases and base editors suitable for treating a wide range of diseases differentiates us in several ways. Now we are focused more on the “bottom-up” approach of discovering and developing CRISPR systems to treat select diseases for our partners and our own therapeutic development programs.

First, we’re not just working with state-of-the-art computation and AI: we’re pushing their boundaries. We know this, because the questions we’re asking some of the technology sector’s AI leaders are stretching their capabilities into new areas.

Second, we have collected and curated our own proprietary protein database, which constitutes a heterogeneous collection of sequences from isolated organisms and metagenomes. The collection contains proteins from large public collections and numerous published studies, along with proprietary sequences from diverse biomes. These data have been unified, annotated, and structured to promote model development toward protein engineering.

Third, we are generating a large body of experimental data, tailored to the therapeutic needs CRISPR aims to address. We’re using these data to build and train AI models that can predict and generate CRISPR systems, customized to the target disease sequences with high activity and specificity. These AI-generated systems should have significantly higher success rates as gene editing therapies than natural systems and could broaden the disease areas treatable with CRISPR.

Finally, the integration of our computational and experimental labs enables a seamless transition for prediction to experimental validation of the AI-generated CRISPR systems. Our structure allows our computational and experimental teams to work hand in hand, framing questions for our AI models in terms of how the output could be tested experimentally. We want to know: Is the model really gaining us something? Is this something we want to do again? We’re testing whether AI-based recommendations are better than human recommendations – something the field believes should be true but hasn’t yet proven generally.

With our AI CRISPR models, Life Edit can serve not only the needs of our own pipeline, but the whole field of CRISPR gene therapy as well through partnerships. With our models and portfolio of validated CRISPR systems, we are the first stop for a company starting a CRISPR gene therapy program; with our enabling technologies and the end-to-end manufacturing capabilities of ElevateBio’s BaseCamp, we are a “one-stop shop” and ideal partner for developing a CRISPR drug.

We can offer partners and clients many flexible options, not just one, for developing and commercializing their gene editing therapies, thus putting more shots on goal as an industry than any one organization could otherwise pursue independently. We are propelling the entire CRISPR gene editing field into the future, to bring these potentially curative medicines to as many patients as possible.

Capabilities & Services

Expertise