EXODEC: A Rational Design Framework for BBB Ligand Evaluation and De Novo Peptide Engineering

Over 98% of neurological drugs fail because they cannot cross the blood–brain barrier. A major reason is that most delivery strategies focus only on binding, even though successful transport requires a sequence of steps: vascular adherence, receptor binding, uptake, transcytosis, and release. Optimizing just one step can actually cause failure such as the in transferrin paradox, strong binding leads to recycling instead of delivery.

To address this, I developed EXODEC, a computational platform for BBB ligand evaluation and de novo peptide design. EXODEC models all five stages of transport and integrates them into a unified scoring system. It then expands into a broader framework including a BBB atlas, digital-twin simulations, safety screening, cross-species modeling, docking, and autonomous peptide generation.

My results showed that binding alone is not a reliable predictor of delivery. Transcytosis correlated more strongly with overall performance, and benchmark ligands like transferrin performed poorly due to downstream failure. In contrast, EXODEC-designed peptides—such as EXODEC_L4 and EXODEC_GEN_001—consistently outperformed established ligands and remained robust across simulated biological conditions.

Overall, EXODEC reframes BBB drug delivery from optimizing a single interaction to engineering performance across the entire transport pathway, providing a more reliable strategy for designing effective brain-targeted therapeutics.

More than 98% of neurological therapeutics fail to cross the blood–brain barrier (BBB), limiting treatment for diseases such as glioblastoma and Alzheimer’s disease. The BBB is a highly specialized, selectively permeable interface formed by brain microvascular endothelial cells , sealed by tight junction proteins (e.g., claudins, occludin, ZO-1) and reinforced by efflux transporters and metabolic enzymes. While these mechanisms preserve neural homeostasis, they also prevent most drugs—especially large or hydrophilic molecules—from reaching the brain.

This project was motivated by a central question: why do so many promising therapeutics fail at the BBB, and how can delivery be improved? Current strategies often optimize only receptor binding, despite the fact that transport occurs through a multi-step process known as receptor-mediated transcytosis (RMT), involving adherence, binding, uptake, intracellular trafficking, and release. Failures at any stage can prevent successful delivery; for example, excessively strong binding can lead to endosomal recycling rather than release, a phenomenon known as the “binding site barrier.”

To address this, I explored a multi-stage framework for evaluating and optimizing ligand-mediated delivery using extracellular vesicles (EVs). EVs are promising carriers due to their biocompatibility and natural role in intercellular transport, but require surface functionalization to achieve BBB specificity.

Improving BBB delivery could benefit patients with currently untreatable neurological diseases by enabling targeted therapies to reach the brain more effectively. More broadly, this work contributes to a shift from single-parameter optimization toward integrated, system-level design, with the potential to accelerate drug development and improve clinical outcomes worldwide.

To move beyond reductionist binding-focused screening, I developed EXODEC, a modular framework that models ligand-mediated BBB delivery across the full transport lifecycle. Each ligand was stored with a metadata.json file and a structure.pdb file, linking its sequence, design rationale, receptor target, payload assumptions, and docking inputs.

Ligand design:

L1 was a benchmark shuttle combining favorable traits from top established BBB ligands.

L2 was derived from LRP1's C3 binding domain and modified with histidine substitutions to create a pH-sensitive binding/release switch.

L3 was designed around the CD98hc structure using receptor geometry.

L4 was based on L3 but optimized to reduce EXODEC penalties; I increased its size and added a triple-histidine pH switch.

I then tested an L4 mutation set by swapping residues and rescoring substitutions, but the original L4 remained the best predicted design.

EXODEC_GEN_001 was created through EXODEC's autonomous generator-evaluator loop, which generated candidates from high-performing motifs, receptor constraints, pH-switch logic, and penalties learned from earlier designs.

Testing and data:

Each candidate was scored from 0.0-1.0 across vascular adherence, receptor binding, cellular uptake, transcytosis efficiency, and parenchymal release, so strong binders could be penalized if they failed downstream transport.

Atlas and digital twin:

Data were collected by screening the library across healthy, inflamed, neurodegenerative, and glioblastoma-associated BBB contexts. The Atlas mapped stage-specific bottlenecks, rank shifts, and context-dependent failure modes, while the digital twin modeled receptor competition, endogenous occupancy, efflux tone, enzyme burden, ligand fate decomposition, and virtual-patient variability to test whether leads stayed robust across simulated biological conditions.

Outputs:

Additional modules assessed off-target safety, cross-species behavior, payload compatibility, pharmacokinetics, and geometry-aware docking. Outputs were integrated into master figures, manuscript tables, docking summaries, and an interactive transcytosis dashboard.

EXODEC expanded from a ligand scorer into a systems-level BBB delivery platform and produced computational results across transport, safety, structure, disease context, and design. Atlas-scale scoring showed that successful BBB delivery was not determined by receptor binding alone (Fig. 1). Across the ligand set, transcytosis and release appeared as major bottlenecks, and many ligands with acceptable binding still ranked lower because recycling or incomplete release reduced transport output (Fig. 2). The BBB transcytosis atlas separated ligands by score and stage-specific bottleneck class (Fig. 1 and Fig. 2).

Within the EXODEC family, EXODEC_L4 remained a strong lead scaffold, while EXODEC_GEN_001 emerged as a highly competitive variant (Fig. 1 and Fig. 4). In downstream analyses including GBM-focused prioritization, payload-aware ranking, and pharmacokinetic simulation, EXODEC_GEN_001 matched or exceeded EXODEC_L4 (Fig. 5). In the glioblastoma module, EXODEC-family ligands ranked above many comparators, and the dominant limitation for top GBM candidates was incomplete transcytosis with recycling rather than receptor failure (Fig. 3 and Fig. 5).

The digital twin showed a key systems-level result: ligand ranking changed across biological states (Fig. 2). Healthy, inflamed, neurodegenerative, and glioblastoma-associated BBB environments did not preserve one universal leaderboard. Receptor crowding, competition, enzyme burden, and altered transport context reshaped ligand performance (Fig. 2 and Fig. 3). The virtual-patient layer identified candidates that remained robust under simulated variability rather than a nominal endothelial state (Fig. 2).

The safety layer added a second ranking axis beyond transport. By integrating delivery with predicted off-target liability, EXODEC generated a combined brain-delivery-and-safety score that reordered candidates relative to transport-only ranking (Fig. 4). Cross-species simulations separated designs with broader species robustness from those with stronger human-specific behavior (Fig. 4). Payload-aware modeling showed that ligand rankings changed across cargo classes, with the best ligands for siRNA or ASO not always matching those for larger biologic payloads (Fig. 5).

The molecular docking layer added a structural dimension. Using local ligand and receptor PDB files, EXODEC scanned receptor pockets and scored geometry-aware ligand-pocket compatibility, producing native-pair docking summaries and a docking gallery (Fig. 5). All outputs were consolidated into a manuscript tables and an interactive dashboard.

Statistics

The regenerated scoring run contains 23 ligand entries and 7,292 atlas-context rows across four BBB states (Fig. 1 and Fig. 2). The top five mc_p10-ranked ligands were EXODEC_L4 (0.692), EXODEC_GEN_001 (0.689), EXODEC_L5 (0.626), UNIVERSAL_SHUTTLE (0.608), and EXODEC_L4_Mutation (0.537) (Fig. 1). Rank-correlation analysis supported the EXODEC premise: transcytosis correlated more strongly with mc_p10 than binding alone (Spearman rho 0.677 vs. 0.528), while release remained strongly associated with performance (rho 0.592) (Fig. 2). This was clearest in the transferrin paradox comparison: Transferrin retained moderate binding (0.599) and release potential (0.844) but scored 0.000 for transcytosis and only 0.195 mc_p10, whereas EXODEC_GEN_001 combined strong binding (0.816), transcytosis (0.675), release (0.981), and mc_p10 (0.689) (Fig. 2). Across the designed ligand set, mean mc_p10 was 0.533, 1.67-fold higher than the established-ligand benchmark mean of 0.320 (Fig. 4).

The development of EXODEC addresses a major problem in neuropharmacology: the persistent failure of affinity-first BBB ligand design. Rather than treating BBB transport as a single docking event, EXODEC models it as a five-stage biological lifecycle. This shifts the design objective from "strongest binder" to "best overall transporter."

That change matters because many CNS therapeutics fail upstream of efficacy. A potent drug is not useful if it cannot reach the brain, and a brain-penetrant ligand is not automatically useful if it is unsafe, species-limited, or trapped in endothelial recycling pathways. By connecting atlas-scale screening, digital twin simulation, safety-aware prioritization, payload-aware modeling, docking, and de novo design in one framework, EXODEC provides a practical way to prioritize stronger candidates and more informative experiments.

At the same time, the project remains computational and hypothesis-generating. The docking layer is geometry-aware but not full atomistic simulation, and the biological modules are grounded in curated assumptions rather than direct wet-lab measurement. For that reason, the most defensible claim is not that EXODEC proves biological truth, but that it improves prioritization, mechanistic insight, and experimental targeting. In that sense, the project represents a transition from trial-and-error discovery toward computational engineering of BBB delivery strategies.

The immediate next phase of this research is a formal collaboration with the Breyne Lab at Harvard Medical School to conduct in vivo validation of EXODEC_GEN_001. Using murine models, we will correlate my framework’s "Transcytosis Scores" with actual brain parenchymal accumulation through fluorescence microscopy. Beyond validation, I plan to expand the DMS Atlas to include other high-traffic targets like the Insulin Receptor (IR). Ultimately, I aim to integrate Graph Neural Networks to predict how larger therapeutic payloads, such as CRISPR-Cas9 complexes, affect the ligand’s ability to navigate the blood-brain barrier.

I would like to extend my deepest gratitude to Dr. Koen Breyne of Harvard Medical School for his invaluable mentorship. His expertise in neuro-immunology and receptor kinetics provided the critical feedback necessary to bridge the gap between my computational models and biological reality. I also want to thank the West Point Grey Academy science department for fostering my curiosity and providing the resources to pursue such a complex project.

To the organizers of the Greater Vancouver Regional Science Fair and Youth Science Canada, thank you for the opportunity to present my work at a national level. Most importantly, I am incredibly grateful to my parents and friends; their encouragement kept me motivated through the thousands of lines of code and the many late nights spent debugging the EXODEC framework. This project would not have been possible without this incredible community of support.

Selected References:

Abbott NJ, Patabendige AA, Dolman DE, Yusof SR, Begley DJ. Structure and function of the blood-brain barrier. Neurobiology of Disease. 2010 Jan;37(1):13–25. doi:10.1016/j.nbd.2009.07.030.

Abbott NJ, Rönnbäck L, Hansson E. Astrocyte-endothelial interactions at the blood-brain barrier. Nature Reviews Neuroscience. 2006 Jan;7(1):41–53. doi:10.1038/nrn1824.

Baghirov H. Mechanisms of receptor-mediated transcytosis at the blood-brain barrier. Journal of Controlled Release. 2025 May 10;381:113595. doi:10.1016/j.jconrel.2025.113595.

Baghirov H. Receptor-mediated transcytosis of macromolecules across the blood-brain barrier. Expert Opinion on Drug Delivery. 2023;20(12):1699–1711. doi:10.1080/17425247.2023.2255138.

Brixi G, Durrant MG, Ku J, et al. Genome modelling and design across all domains of life with Evo 2. Nature. 2026. https://doi.org/10.1038/s41586-026-10176-5

Daneman R, Prat A. The blood-brain barrier. Cold Spring Harbor Perspectives in Biology. 2015;7(1):a020412. doi:10.1101/cshperspect.a020412.

Guo K, Cao D, Marchese-Thomas LP, Dong Y. Antibody engineering for receptor-mediated transcytosis across the blood-brain barrier. Bioconjugate Chemistry. 2025;36(10):2109–2115. doi:10.1021/acs.bioconjchem.5c00379.

Haqqani AS, Bélanger K, Stanimirovic DB. Receptor-mediated transcytosis for brain delivery of therapeutics: receptor classes and criteria. Frontiers in Drug Delivery. 2024;4:1360302. doi:10.3389/fddev.2024.1360302.

Kang Z, Zeng C, Tian L, et al. Transferrin receptor targeting segment T7 containing peptide gene delivery vectors for efficient transfection of brain tumor cells. Drug Delivery. 2022;29(1):2375–2385. doi:10.1080/10717544.2022.2102696.

Keshtkar S, Azarpira N, Ghahremani MH. Mesenchymal stem cell-derived extracellular vesicles: novel frontiers in regenerative medicine. Stem Cell Research & Therapy. 2018;9(1):63. doi:10.1186/s13287-018-0791-7.

Leandro K, Rufino-Ramos D, Breyne K, et al. Exploring the potential of cell-derived vesicles for transient delivery of gene editing payloads. Advanced Drug Delivery Reviews. 2024;211:115346. doi:10.1016/j.addr.2024.115346.

Lundsgaard CC, Gbyl K, Videbech P. Blood-brain barrier permeability and electroconvulsive therapy: a systematic review. Acta Neuropsychiatrica. 2023;37:e22. doi:10.1017/neu.2023.48.

Qiu B, Pompe S, Xenaki KT, et al. Receptor-mediated transcytosis of nanobodies targeting the heparin-binding EGF-like growth factor in human blood-brain barrier models. Journal of Controlled Release. 2025;383:113852. doi:10.1016/j.jconrel.2025.113852.

Rayamajhi S, Aryal S. Surface functionalization strategies of extracellular vesicles. Journal of Materials Chemistry B. 2020;8(21):4552–4569. doi:10.1039/d0tb00744g.

van Niel G, D’Angelo G, Raposo G. Shedding light on the cell biology of extracellular vesicles. Nature Reviews Molecular Cell Biology. 2018;19(4):213–228. doi:10.1038/nrm.2017.125.

Zhao Y, Gan L, Ren L, et al. Factors influencing the blood-brain barrier permeability. Brain Research. 2022;1788:147937. doi:10.1016/j.brainres.2022.147937.