AI-enabled discovery and biochemical optimization of minibinders targeting cancer cell-surface proteins

Sequence and structural determinants of efficacious de novo chimaeric antigen receptors.

Advances in generative protein design using artificial intelligence (AI) have enabled the rapid development of binders against heterogeneous targets, including tumour-associated antigens. Despite extensive biochemical characterization, these novel protein binders have had limited evaluation in candidate therapeutics, including chimaeric antigen receptor (CAR) T cells. Here we synthesize generative protein design workflows to screen 1,758 newly designed protein binders targeting BCMA, CD19 and CD22 for efficacy in scalable protein-binding, T-cell activation and in vivo killing assays. We characterize three main challenges that hinder the utility of de novo protein binders as CARs, including tonic signalling, occluded epitope engagement and off-target activity. We develop computational and experimental heuristics to overcome these limitations, including screens of sequence variants of individual parental structures, that retain on-target CAR activation while mitigating liabilities. Together, our framework accelerates the development of AI-designed proteins for future preclinical therapeutic screening, helping enable a new generation of cellular therapies.

Profiling Protein‐Peptide Interactions by Yeast Surface Display

A protocol for discovering protein‐binding peptides using a very large, target‐agnostic yeast surface display library containing approximately 6.1 × 109 unique clones and providing broad coverage of short peptide sequence space is described.

A Mammalian High-Throughput Screen for AI-Designed Peptide-Guided Protein Degraders

Targeted protein degradation (TPD) offers a route to eliminate disease-driving proteins that remain inaccessible to conventional inhibitors. However, degrader discovery remains low-throughput, labor-intensive, and dependent on randomized libraries or non-human display systems, limiting functional selection in mammalian cells. Here, we present a high-throughput, human cell-based platform for screening peptide-guided ubiquibodies (uAbs). These genetically encodable, doxycycline-inducible degraders fuse peptide guides generated by protein language models to the CHIPΔTPR E3 ligase domain, creating a modular, CRISPR-like system for programmable TPD. For each target, we introduce a pooled uAb library into the corresponding fluorescent reporter cell line, isolate cells with reduced target abundance by FACS, and recover enriched peptide guides by sequencing. For β-catenin, enriched uAbs reduced endogenous β-catenin abundance and Wnt signaling in DLD1 cells. GFAP-directed uAbs reduced endogenous GFAP abundance and cell viability in U251 glioblastoma cells, while EWS::FLI1-directed uAbs reduced fusion oncoprotein abundance, suppressed EWSAT1 expression, and increased apoptosis in Ewing sarcoma models. Finally, a screen using endogenously tagged GATA2 further identified uAbs that reduced GATA2 under native genomic regulation. Overall, our platform connects generative peptide design to functional mammalian selection and establishes a scalable strategy for CRISPR-like proteome perturbation.

Structure-guided identification and experimental validation of NAT10-targeting small molecules in colorectal cancer cells

Artificial Intelligence-Aided De Novo Design of High-Affinity Protein Binders Against Shiga Toxin 2

Shiga toxin-producing Escherichia coli (STEC) is a major food-borne pathogen with limited treatment options. Targeting Shiga toxin (Stx) represents a promising therapeutic strategy. Using artificial intelligence (AI)-assisted de novo protein design combined with bacterial surface display screening, pull-down assays, and biolayer interferometry, we identified protein binders against the Stx2a B subunit (Stx2aB), among which Binder506 exhibited high affinity. Structural modeling predicted a three-α-helix fold, consistent with circular dichroism analysis, while differential scanning fluorimetry revealed high thermostability. Importantly, Binder506 demonstrated potent Stx2a-neutralizing activity with minimal intrinsic cytotoxicity in Vero cells. Size-exclusion chromatography and coimmunoprecipitation showed that Binder506 did not disrupt Stx2a holotoxin integrity. Instead, cellular binding and internalization assays demonstrated that Binder506 inhibited Stx2a attachment to and uptake by Vero cells. Collectively, these findings demonstrate the potential of AI-assisted de novo protein design for generating effective Stx-neutralizing proteins and establish a promising framework for the development of novel therapeutics.

Bridging AI, molecular dynamics, and experiments for mechanism-aware TCR engineering in cancer immunotherapy

T cell receptor engineering holds great promise for cancer immunotherapy, yet the identification and optimization of high-affinity, functionally effective TCRs remain challenging. While advances in artificial intelligence have enabled large-scale exploration of TCR sequence space, these approaches often lack mechanistic interpretability and generate extensive candidate pools that are difficult to prioritize. Conversely, physics-based methods such as molecular dynamics simulations provide detailed insight into binding energetics and conformational dynamics but are not suited for high-throughput screening. Here, we propose an integrated, mechanism-aware framework that bridges AI-driven design, MD-based refinement, and experimental validation. In this strategy, biologically informed selection of initial TCRs is followed by AI-driven generation of variants, while molecular dynamics simulations act as a low-throughput, high-information filter to uncover the molecular determinants of TCR-pMHC interactions, including allosteric communication. Importantly, MD is positioned not only as a selection tool but also as a generator of mutation hypotheses. Experimental validation using surface plasmon resonance and functional cellular assays such as RAPTR enables the identification of TCRs that are both high-affinity and functionally active. This integrated approach establishes a closed-loop, multi-scale design paradigm that combines exploration, mechanistic understanding, and functional validation, providing a promising strategy for the rational development of therapeutic TCRs.