Rakovina Demonstrates AI-Designed Brain-Penetrant Dual ATR-mTOR Inhibitor

Rakovina Therapeutics presented new preclinical data at the AACR 2026 Annual Meeting showing a generative-AI-designed, brain-penetrant dual ATR-mTOR inhibitor with demonstrated in vivo efficacy in PTEN-deficient tumor models, and reporting successful characterization of a patterned lipid nanoparticle (LNP) formulation for its bifunctional PARP/HDAC candidate, kt-3283. The work was developed using the Enki generative AI platform in collaboration with Variational AI and was presented in the DNA Damage and Repair sessions.

The dual inhibitor program targets two survival pathways, ATR and mTOR, in PTEN-deficient cancers, a vulnerability linked to loss of genomic stability and activated survival signaling. The team used a latent diffusion model in Enki to co-optimize multiple objectives including potency, selectivity, central nervous system penetrance, and ADMET properties. Poster-level disclosures include efficacy and pharmacokinetic readouts in PTEN-deficient models and evidence of brain penetrance.

• •Simultaneous inhibition of ATR and mTOR in a single small molecule scaffold
• •Optimization for CNS penetrance to address brain-metastatic and primary brain tumors
• •Demonstrated in vivo efficacy and supporting PK data for the lead compound

kt-3283 and delivery advances: kt-3283 is a bifunctional molecule that inhibits PARP and HDAC, aiming to replace combination regimens that drive additive toxicity. Rakovina presented in vitro potency across multiple tumor types and described successful encapsulation into patterned LNPs. The LNP formulation was characterized for stability and payload incorporation, positioning it as a route to improve biodistribution and reduce systemic toxicity associated with dual-agent regimens.

These results are notable because they combine two accelerating trends: generative-AI-driven medicinal chemistry and advanced delivery systems. Using Enki to navigate a high-dimensional optimization problem enables simultaneous tradeoffs that are otherwise costly in iterative chemistry campaigns. PTEN loss is prevalent in hard-to-treat indications, occurring in up to 40% of gliomas and 63% of certain breast cancers, so a brain-penetrant ATR-mTOR strategy addresses an unmet clinical niche. The kt-3283 program tackles the practical barrier that effective combinations often fail clinically because of overlapping toxicities; a single-molecule bifunctional approach plus LNP delivery is a pragmatic engineering solution.

For ML and drug-discovery teams this is a concrete case where a generative model accelerated candidate selection and property balancing across potency, selectivity, CNS exposure, and ADMET. For translational biologists and formulation scientists the LNP work shows how delivery engineering remains essential even when molecules are optimized computationally. The combination highlights an R&D pattern: use AI to produce better molecules, and use formulation to make those molecules deployable in vivo.

Verify reproducibility and translational robustness, particularly CNS exposure across species and safety margins for the dual ATR-mTOR inhibitor. For kt-3283, watch for in vivo efficacy and tolerability data with the LNP formulation and any IND-enabling toxicology plans. Continued disclosure of assay protocols, PK metrics, and model training constraints will determine how broadly applicable these generative-AI workflows are to other targets.