Antibody diversity is traditionally viewed as arising from combinatorial heavy-light chain pairing, yet the extent to which this process is truly random remains unclear. I will present ImmunoMatch, a language-model–based framework that learns the sequence determinants of heavy-light chain compatibility. By uncovering the molecular grammar of antibody assembly, ImmunoMatch provides a quantitative compatibility score with applications in antibody discovery, repertoire analysis, and the design of next-generation of antibody therapeutics.

De novo antibody design is maturing, yet uneven. I’ll share lessons from multiple campaigns—hit rates, epitope-to-epitope variability, and methodological biases—anchored by the critical need for standardized benchmarks. We’ll examine metrics: which truly predict success and which mislead. I’ll highlight inflection points for success rate and emphasize needs beyond binding—robust expression and minimized aggregation—to translate computational promise into developable therapeutics.

Gain-of-function defects in sodium channel (Na_V) function that amplify late/persistent sodium current (I_NaL) are associated with multiple human diseases, including cardiac arrhythmia and epilepsy. Despite intense effort, engineering synthetic modulators that selectively inhibit I_NaL remains extremely challenging. Here, we have used a computational platform to engineer a de novo peptide modulator, ELIXIR, that targets elevated I_NaL. We show that ELIXIR selectively inhibits ‘pathogenic’ I_NaL and confirm its functionality in cellular and murine models of cardiac arrhythmia and epilepsy. Our findings highlight the potential utility of de novo protein design in the rational development of synthetic ion channel modulators.

When we engineer proteins for human applications, it is common to simultaneously optimize for multiple objectives. Such tasks are cumbersome or even impossible with conventional experimental approaches. I will share two ML/AI-enabled examples, one to create antibody-like binders, and the other to remove potentially immunogenic peptides from biologics.

Adaptive immune receptor repertoires encode a history of antigen exposure and disease, making them attractive targets for machine-learning (ML)-based diagnostics and therapeutic discovery. However, the absence of standardized benchmarks has hindered objective evaluation of competing methods. We organized the first community challenge to benchmark ML approaches on two central repertoire-analysis tasks: predicting immune state from labelled repertoires (a diagnostic use case) and recovering immune-state-associated receptors (a therapeutic discovery use case), using approximately 75,000 experimentally generated and biologically realistic simulated T-cell receptor repertoires. Across 20 assessed methods, predictive performance remained modest, with the best-performing approaches achieving average ROC AUC values of only 0.70–0.75. Recovery of immune-state-associated receptors proved substantially more challenging, with the top method achieving a Jaccard similarity of approximately 0.15 (range 0-1). By systematically dissecting how performance varied across datasets and study designs, the challenge identifies key methodological limitations, establishes a community benchmark for immune repertoire ML, and provides a roadmap for future advances in ML-based immunodiagnostics and therapeutic discovery.

The Immune Epitope Database (IEDB) curates experimentally defined immune epitopes and their associated T and B cell receptors, currently comprising ~185,000 TCRs and ~5,000 antibodies with validated specificity. We applied a standardization and validation pipeline to the full receptor dataset, correcting nomenclature and sequence delimitation inconsistencies. The resulting harmonized dataset is ready to use for bioinformatics applications, including immune repertoire annotation and machine learning-based receptor specificity prediction.

AF3-TurboAb is a scalable framework for repertoire-scale antibody–antigen complex structural decoding and structure-guided antibody engineering. By removing preprocessing bottlenecks, AF3-TurboAb enables rapid, end-to-end complex modeling while preserving near-experimental interface fidelity. Applied to large immunization-derived nanobody repertoires, it reveals unmapped epitopes, affinity hotspots, and structurally convergent binding solutions across diverse sequences. We demonstrate its utility for designing durable multiepitope neutralizers against highly evolved viruses, identifying cross-species and glycoform-specific cancer checkpoint binders, enabling near-real-time in silico binder triage, and supporting multivalent and multispecific engineering for challenging biomedical applications.

The paradigm of biologics discovery is fundamentally shifting from discovering what exists in nature to designing what is required in the clinic. This presentation explores how generative AI enables the de novo creation of novel therapeutic proteins with tailored functionalities, discussing the current breakthroughs and future landscape of this revolution.

Achieving precise control over cell surface machinery requires approaches that extend beyond existing ligands. De novo protein design allows targeting virtually any epitope on a given protein. We designed de novo proteins that engage endogenous internalizing receptors for cell-type-specific degradation and delivery of protein-based nanocages. To further enhance the precision of designed proteins, we sought to program selective responsiveness to native environmental cues that define cellular state and location. Together, these efforts establish a foundation for employing molecular engineering and ML-driven protein design to create precision molecules that reprogram cellular processes.