The competition’s premise deliberately raised the bar beyond standard binder benchmarks. A CAR binding domain must do far more than dock onto its antigen: it must be synthesized as DNA, expressed on the T cell surface within a fixed receptor scaffold, and trigger signaling that is strong enough to activate the cell but not so strong that the T cell becomes exhausted. Teams were told their 80-amino-acid binders would replace the single-chain variable fragment, or scFv, of a second-generation CAR28z construct, a domain that in natural antibodies typically runs 230 to 260 amino acids. Success ultimately depended on CAR-T cell function, not just molecular stickiness. Organizers provided primers on CAR-T biology, the exact scaffold, and assay details, then let competitors loose with whatever AI-driven workflows they preferred, creating an unusually diverse and honest snapshot of how the field actually designs proteins today.

The methodological landscape that emerged was dominated by generative diffusion models. Eighteen teams used RFdiffusion, two used sequence-structure co-design tools such as Chroma, and others turned to hallucination-based approaches like BindCraft, EvoBind, and ColabDesign, or protein language models including PepMLM, ESM-2, and RayGun. Nearly twenty teams employed ProteinMPNN or its soluble variant to design sequences onto fixed backbones, and seventeen used AlphaFold2 to verify that their creations were predicted to fold as intended. Twelve teams essentially followed the default RFdiffusion-ProteinMPNN-AlphaFold2 pipeline, while nine practiced iterative diffusion, trimming low-scoring segments and regenerating until scores improved. The result was extraordinary sequence diversity: more than 83 percent of submissions showed no significant alignment to any sequence from another team, although within-team redundancy was high, with a fifth of sequences nearly identical to a teammate’s design.

Testing 12,000 designs individually would have been impractical, so the organizers built a pooled screening system with elegant internal bookkeeping. Each codon-optimized binder sequence was cloned into the CAR plasmid library, where the binder itself served as a unique DNA barcode. Using the GeneWeld targeted integration method, the library was knocked into the T cell receptor alpha chain locus of primary human T cells, generating a pool of TRAC-negative, CAR-positive cells. After methotrexate selection using a co-expressed DHFR variant, the cells were split: one arm cultured alone, the other challenged repeatedly with mitotically inactivated CD20-positive Raji tumor cells. Over two weeks, binders that enabled productive antigen recognition drove proliferation of their host cells, and next-generation sequencing of the barcode region revealed which designs expanded. Comparing tumor-challenged cultures against the no-target controls allowed antigen-dependent growth to be quantified robustly for every design simultaneously.

The screen produced a sobering attrition curve. Nearly 98 percent of designs passed DNA synthesis quality control, but only 56.8 percent were recovered at sufficient sequencing depth after growth, implying that many binders failed to yield a viable CAR, likely because of poor expression, folding, or membrane localization. Ultimately, 707 designs, just under 6 percent, showed significant CD20-specific enrichment of at least twofold. Hit rates varied dramatically between teams, from 0.6 percent to 38.4 percent, a spread that itself encodes lessons about which strategies work. The ten best non-redundant designs advanced to individual validation, where each was rebuilt as a standalone CAR construct and subjected to a full battery of functional assays against CD20-positive Raji cells and CD20-negative K562 controls.

Seven of the ten leading designs outperformed untransfected control T cells in proliferation, expansion, and cytokine production, with flow cytometry confirming CAR expression in most of them. Designs 1506, 5300, and 5981 were the strongest cytokine producers, generating as much interleukin-2 and interferon-gamma as the scFv positive control, and design 5300 exceeded it. When it came to the most demanding readout, specific killing of CD20-positive targets, only four designs, 1506, 2383, 3494, and 3718, achieved significant target-specific lysis above background, and two of those matched the scFv control in specificity. Notably, several broadly functional designs showed no measurable binding in isolation, suggesting they may have relied on avidity or multivalent mechanisms within the receptor context. The pooled hit rate of 707 thus appears to be an upper bound on true functional success.

Surface plasmon resonance against detergent-solubilized CD20 added a crucial biophysical dimension. The scFv control bound with a dissociation constant of 1.75 nanomolar, but only three of the top ten designs showed appreciable binding: design 2383 at 643 nanomolar, and designs 1506 and 5981 with weak but detectable signals between roughly 200 and 600 nanomolar. The authors caution that this apparent preference for moderate or weak affinity is consistent with CAR biology, where excessively tight binding can drive trogocytosis and T cell exhaustion, and lower-affinity receptors have sometimes eliminated tumors better. However, the binding assay tested isolated 80-mers produced by bacterial cell-free synthesis rather than full receptors in mammalian cells, so some failures may reflect structural context rather than design quality. The team recommends that future efforts fold candidate binders within the complete CAR sequence, a computationally trivial step that only one team actually performed, and that team produced the top-performing design.

The most consequential findings came from mining the entire dataset for predictors of success and failure. The organizers computed more than 400 features per sequence, from GC content and Shannon entropy to predicted secondary structure and interface energies. Strikingly, 98.9 percent of designs that vanished from the proliferation screen had been created with ProteinMPNN or SolubleMPNN, and the failed designs were almost pure alpha helices, averaging 83 percent helical content, densely packed with lysine and glutamate. The ratio of residues in lysine-plus-glutamate helices was by far the strongest predictor of recovery failure, reaching 0.91 ROC-AUC in a cross-validated logistic model, averaging 35 percent of residues in failed designs versus 15 percent in recovered ones. The likely mechanism is translational interference: glutamate repeats are known to stall ribosomes, and lysine and glutamate codons produce adenosine-rich DNA, with poly-A tracks also disrupting translation. Failed sequences carried nearly four glutamate-glutamate repeats on average versus 1.6 in recovered designs, and adenosine dinucleotide repeats predicted recovery with 0.88 ROC-AUC. An independent phage display study has since confirmed the lysine-glutamate association, suggesting the signal generalizes well beyond CAR-T cells.

Other failure modes emerged with equal clarity. Designs that caused CD20-specific depletion of T cells, possibly reflecting exhaustion, were enriched in cysteines predicted to form intramolecular disulfide bonds, which may have exposed hinge sulfhydryls and amplified tonic signaling. DNA synthesis failures clustered among sequences with high GC content and low entropy. Perhaps most humbling for the structure-prediction community, the standard confidence metrics, including ipTM, pLDDT, ipSAE, LIS, SAP, PDockQ, and language-model likelihoods, showed little predictive power for functional outcomes, and most teams’ own confidence rankings correlated poorly, or even negatively, with experimental success. Simple sequence statistics beat sophisticated structure-based scores. Applying retrospective filters for GC content, entropy, adenosine repeats, glutamate repeats, and lysine-glutamate helix content removed 4,600 sequences and nearly doubled the proportion of designs showing CD20-specific proliferation, from 5.9 to 7.6 percent, rising to 10.6 percent with a cysteine filter that would also have eliminated two of the non-functional top-ten designs.

The authors are careful about the study’s limits: a single CAR scaffold, a single donor isolate, a comparison of Raji-challenged cells against no-target rather than CD20-negative controls, and a design space constrained to 80 amino acids all shape the conclusions. No design is remotely clinic-ready, and all underperformed the scFv control in most readouts. Yet the significance of the work lies less in any single binder than in the demonstration that generative protein design can now be evaluated end-to-end in a therapeutic context, at scale, in the open. The dataset has been released on Zenodo with code on GitHub, and the practical recommendations, avoid lysine-glutamate helix overload, watch for cysteines, fold within the full receptor, and treat moderate affinity as a feature rather than a bug, offer the field its first empirically grounded playbook for designing proteins that do not merely bind, but function inside living cells.

Subject of Research: Experimental validation of AI-designed de novo protein binders as CAR-T cell recognition domains targeting CD20

Article Title: Validation and analysis of 12,000 AI-driven CAR-T designs in the Bits to Binders competition

Article References: Kosonocky, C. W., Abel, A. M., Feller, A. L., Cifuentes Rieffer, A. E., Woolley, P. R., Lála, J., Barth, D. R., Gardner, T., Bits to Binders Competitors, Acosta, D., Angioletti-Uberti, S., Annapure, V., Anthony, N., Barghout, R. A., Beining, M., Bibekar, P., Biton, D., Bryant, P., Bushuiev, A., … Marcotte, E. M. (2026). Validation and analysis of 12,000 AI-driven CAR-T designs in the Bits to Binders competition. Molecular Systems Biology. https://doi.org/10.1038/s44320-026-00246-1

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00246-1

Keywords: CAR-T cells, protein design, artificial intelligence, CD20, ProteinMPNN, RFdiffusion, de novo binders, pooled screening, protein engineering, immunotherapy, translational stalling, benchmarking

The research, summarized in a companion Research Briefing, combines AI-assisted de novo binder design with the assembly of complete CAR constructs and scalable testing platforms in vitro and in vivo. Rather than asking whether a computationally designed protein can bind a target antigen in a dish, the investigators pushed the designs through the full developmental gauntlet that a therapeutic candidate must survive: expression on the surface of primary T cells, signal transduction upon antigen engagement, target cell killing, and durable anti-tumor activity in living models. The resulting dataset links sequence-level and structural-level features of designed binders directly to therapeutic function, providing a design grammar that the field has previously lacked.

The work builds on a remarkable run of advances in computational protein design. In 2023, the introduction of the RFdiffusion framework dramatically raised the success rate of de novo binder campaigns by using diffusion models to generate protein backbones tailored to a desired target surface. A year later, researchers reported targeting overexpressed antigens in glioblastoma using CAR T cells armed with computationally designed high-affinity protein binders, offering early proof that AI-designed recognition domains could function as the business end of a chimeric antigen receptor. More recently, the hallucination-based design pipeline BindCraft described one-shot generation of functional protein binders, and it produced some of the highest-success design campaigns in the new study. Together with ProteinMPNN, a robust deep learning method for assigning amino acid sequences to designed backbones, these tools have made binder generation almost routine. The new work addresses the harder question: which of the many binders that pass computational filters will actually drive T cells to kill cancer?

That question matters because binding is only the beginning. A CAR binding domain operates in a demanding mechanical and biological context. It must fold correctly and traffic to the cell membrane when fused to hinge, spacer, transmembrane and signaling modules. It must bind antigen with an affinity that falls within a productive range; too little affinity produces no signal, while excessive affinity can cause antigen-independent tonic signaling, activation-induced cell death and poor persistence. It must tolerate epitope densities that vary enormously across tumor tissues. And it must be small, stable and non-immunogenic enough to be clinically deployable. Natural antibody fragments often fail several of these criteria simultaneously, which is why the pharmaceutical industry has invested heavily in screening campaigns that yield a single usable binder after months of labor.

By systematically varying binder sequences and structures within the CAR context and evaluating the resulting constructs in standardized cell models, the authors were able to infer the attributes that correlate with efficacy. The study frames a complete framework for designing efficacious CARs, in which binder attributes such as affinity, epitope choice, stability and expression behavior are treated as tunable design parameters rather than accidents of discovery. The significance of this reframing is difficult to overstate. For two decades, CAR engineering has been as much an art as a science, with laboratories borrowing fragments from existing antibodies and adjusting hinges and spacers empirically. A predictive model of what makes a binding domain efficacious turns CAR design into an engineering discipline in which candidate receptors can be specified computationally before a single experiment is run.

The experimental architecture underpinning the study is as important as its findings. The investigators paired binder design with scalable in vitro assays that measure how many of the designed constructs express on T cells, how strongly they signal, and how effectively they eliminate antigen-positive targets. In vivo models then tested whether promising designs retained activity against tumors in a physiological setting, where antigen density, immune suppression and trafficking barriers conspire to defeat otherwise potent receptors. This multi-tier funnel mirrors the path of a therapeutic candidate and ensures that the design rules extracted from the data reflect true clinical relevant properties, not merely binding measurements from immobilized proteins.

The implications extend well beyond one cancer type. Because the pipeline is generative, it is in principle antigen-agnostic: given a target surface, the same design-and-test cycle can produce panels of candidate binding domains against antigens relevant to solid tumors, autoimmune disease, fibrosis and infectious disease. Solid tumors have proven especially refractory to CAR therapy, in part because widely shared tumor-associated antigens are also expressed on essential healthy tissues and because single-antigen targeting invites escape. AI-designed binders, selected with precise affinity windows and epitope specificity, could enable new strategies such as affinity tuning to discriminate between high- and low-expressing tissues, dual-antigen logic gating, and rapidly generated panels against patient-specific neoantigens. The glioblastoma work from 2024 demonstrated that computationally designed binders could target an antigen overexpressed in one of the deadliest solid tumors; the new study supplies the general principles for making such binders reliably efficacious.

There are also cautionary notes that seasoned observers of the CAR field will appreciate. Computational design success rates, even with state-of-the-art tools, remain probabilistic, and the attributes that make a binder effective in a standardized cell line may not transfer directly to the hostile microenvironment of a human tumor. Immunogenicity of non-human-derived protein scaffolds must be assessed rigorously before clinical translation, and the regulatory pathway for wholly synthetic recognition domains is still being defined. Tumor heterogeneity, antigen loss and the immunosuppressive microenvironment remain problems that no binder, however well designed, can solve alone. What the study offers is not a finished therapy but a reproducible methodology for generating and selecting binding domains with predictable properties, which removes one of the largest sources of variability and failure in current CAR programs.

The broader scientific community has taken notice of how quickly the ingredients of this advance came together. RFdiffusion and ProteinMPNN provided the generative backbone and sequence design machinery; AlphaFold-style structure prediction supplied reliable in silico validation of designed conformations; BindCraft demonstrated that hallucination-based pipelines could deliver functional binders in single campaigns; and the earlier glioblastoma CAR study established clinical feasibility. The new research closes the loop by asking what distinguishes the binders that work in a CAR from those that bind beautifully on paper but fail on the cell surface. The answer, encoded in the sequence and structural determinants the authors report, is a practical toolkit for the next generation of synthetic immunology.

If the field can standardize on these design attributes, the consequences could be transformative. Cell therapy developers could move from years of empirical binder discovery to weeks of computational specification followed by targeted validation. Clinicians could obtain CAR constructs tuned precisely to the antigen expression profile of an individual tumor. Academic laboratories with modest resources could design receptors against orphan antigens that no commercial entity would ever fund an antibody campaign for. The convergence of generative AI and cellular immunotherapy has promised exactly this kind of acceleration for several years, and this study provides some of the clearest evidence yet that the promise is becoming an operational reality, one amino acid at a time.

Subject of Research: AI-assisted de novo design of protein binders for constructing efficacious chimeric antigen receptor T cell therapies

Article Title: Defining attributes of effective binders for AI-assisted CAR design

Article References: Defining attributes of effective binders for AI-assisted CAR design. (2026). Nature Biomedical Engineering. https://doi.org/10.1038/s41551-026-01792-7

Image Credits: AI Generated

DOI: 10.1038/s41551-026-01792-7

Keywords: CAR T cells, chimeric antigen receptor, AI protein design, de novo binders, RFdiffusion, ProteinMPNN, BindCraft, cancer immunotherapy, protein engineering, Nature Biomedical Engineering, antigen targeting, synthetic immunology