This presentation focusses on two upcoming strategies facilitating CAR T cell generation. Short-term CAR T cells accelerate production times but may bear an increased risk for severe cytokine release syndrome. In vivo CAR T generation relies on T cell-specific vectors as off-the-shelf product while proof for clinical benefit remains to be provided.

Extrahepatic targeting for genetics medicines remains a key challenge. Mirai Bio has built an open platform driven by machine intelligence and in vivo barcoding to develop next-generation ionizable lipids and formulations. These base LNPs can be tuned for high on-target delivery to chosen cell types and favorable off-target selectivity, specifically reduced liver delivery. Further increases in potency and selectivity are seen through the addition of targeting moieties to the surface. Our generative design of delivery solutions can encompass a wide array of specialized nucleic acids and co-optimized with best-in-class manufacturing to develop the future of genetic medicines.

Tonic signaling from CAR expression is proposed to be associated with CAR T exhaustion. Our study describes an in vitro model for investigating tonic signaling in mRNA CAR T cells, which has not been fully characterized. Among approximately 80 tested permutations of structural elements, a few optimal CAR designs showed improved antigen-dependent T cell immune responses in vitro. Additionally, several formats of mRNA were also evaluated for CAR expression persistence.

To improve the efficiency of design and development of CAR constructs, we built the APEX model by fine tuning protein-language model ESM2 with a dataset comprised of binder (scFv, Fab, and VHH) and antigen pairs curated from the Structural Antibody Database. The model achieved robust performance on affinity prediction and can be expanded for other property prediction. Model adoption into CAR campaign facilitates binder prioritization to reduce discovery cycle time.

We present a streamlined workflow integrating machine learning-based protein design, outsourced lineal dsDNA synthesis, and rapid cell-free protein expression. By measuring binding kinetics directly from crude expression mixtures, we obtain kinetic binding data within 5 days of design. We have validated kinetics measurements from crude cell free mixtures, compared to purified proteins, showing strong correlation and reliability. This modular platform enables high-throughput screening and optimization of miniprotein binders with unprecedented speed and efficiency.

• Dive into the challenges with engineering immune cell engagers where machine learning might have an impact (i.e. development speed, therapeutic efficacy, clinical safety)
• Discuss the data and models needed to develop superior therapies. 
• Highlight different machine learning methods being applied to the design and optimisation of immune cell engagers. 
• Explore future directions for machine learning in immune cell engagers. 

Here, we present our lab's recent work on developing programmable protein binders via generative artificial intelligence for targeting heterogenous cancer antigens.

T cells recognize intracellular targets presented by HLA to enable potent anti-tumor immune responses, and these targets can be leveraged to generate off-the-shelf therapeutics using T cell-bispecific engagers to treat a broad patient population. We've developed 3T-TRACE to rapidly identify the antigens of orphan T cells from patient tumors and 3T-PRIME, a TCR mimetic platform to rapidly generate potent and specific binders for therapeutic development.

Immuno-STATs represent a Fc fusion molecules capable of engaging, activating, and amplifying disease-specific T-cells. Our leading clinical candidate, CUE-101, specifically targets HPV E7-specific T cells and has shown remarkable efficacy in patients with advanced head and neck cancer. With this clinical success and our platform's proven safety, we introduce the Immuno-STAT platform. It enables the targeting of various tumor types in addition to AI applications.

A novel conditionally activating costimulatory T cell engager (TCE) platform demonstrates strong potential for autoimmune and solid tumor indications. A 41BB costimulation built-in CD19 x CD20 TCE achieved CAR-T–like potency and superior efficacy over benchmark TCEs in autoimmune models. Similarly, a CD2-costimulated DLL3 TCE outperformed Tarlatamab™ in SCLC models. Conditional activation upon antigen engagement mitigates systemic cytokine release, offering a safer, highly effective strategy for next-generation TCE development.

Oncolytic viral therapies for cancers are frequently limited to intralesional injection due to non-specific localization of systemically injected viral particles. We developed a modified vesicular stomatitis virus (VSV) that harbors tumor-activating moiety. This novel virus preferentially distributes to tumor tissues and showed enhanced safety compared to wild-type VSV. When incorporated with a single chain, biologically active interleukin-12 (IL-12)—the novel virus mounts effective control against tumor growth.