This presentation will focus on novel, chemically enhanced phage libraries. Phage display is a high-throughput screening platform for peptide discovery, which has, however, been limited to natural peptides. Chemical modification of phage, via novel and highly efficient chemistries, allows access to nonnatural peptide libraries and greatly expands the power of this high-throughput screening technology towards the development of peptide therapeutics.

In vivo phage display enables peptide discovery within native environments, but the large size of phage particles limits extravasation into target tissues, resulting in underrepresentation of relevant binders. Comparative in vivo and ex vivo phage display across fresh tissues defines how tissue accessibility shapes peptide recovery and selection bias. These differences are critical for peptide delivery and for evaluating off-target binding that may be underestimated by extravasation constraints.

Macrocyclic peptides have gained increased attention amid claims they are a “Goldilocks” therapeutic modality that can encode the selectivity of a biologic in a footprint close to that of a small molecule. Here we attempt to find a peptide that binds selectively to NRAS, sparing HRAS and KRAS, while accessing the cytosol via passive cell permeability. To do so, we combine subtractive affinity selection with mRNA display to identify Compound 1, an 11mer macrocyclic peptide which binds NRAS at a novel allosteric site between Helix 3 and Helix 4 of the GTPase domain. Compound 1 has total isoform selectivity and can be tuned to achieve activation-state selectivity with a single amino acid change. While it has preferential affinity for oncogenic NRAS specific mutations, it does not inhibit NRAS function or achieve passive membrane permeability.

In this work, we developed a novel pipeline for engineering de novo miniproteins to target the Interleukin-4 receptor alpha (IL4Ra), a crucial receptor in Th2-mediated inflammatory diseases currently dominated by monoclonal antibodies. To achieve this, we used a scaffolding strategy to extract key cytokine helices and used RFdiffusion to design the connecting loops. Initial binders with sub-micromolar affinities were then optimized using molecular dynamics and phage display libraries for affinity maturation. Ultimately, this combined generative AI and directed evolution approach yielded high-affinity miniprotein binders, proving that diffusion models can successfully replicate complex cytokine interfaces for next-generation therapeutics.

Transcription factors and other intracellular protein-protein interactions remain challenging therapeutic targets. This presentation describes an integrated live-cell discovery platform combining genetically encoded peptide libraries, intracellular cyclisation, functional selection and scalable hit validation. By constraining peptides during selection, the workflow enriches cell-tolerated, biostable antagonists that disrupt transcription factor-DNA and protein-protein interactions. The approach links library design, screening, sequencing-based prioritization, biophysical characterization, and cellular validation to accelerate discovery of functional, cell-active constrained peptides.

Peptide-based delivery vectors offer superior performance in targeted radiopharmaceuticals (TRP) when compared to large antibody molecules. One of the bottlenecks in TRP is rapid access to potent and specific peptide vectors that bind to desired cell surface receptors. My talk will describe an integrated technology for discovery of peptide-based TRP vectors that employ chemically enhanced phage display, machine learning from DNA-encoded data, and iterative molecular optimization that integrates both approaches.

By integrating approximately 200,000 mass spectrometry datasets with ribosome profiling and immunopeptidomics resources, we identified more than 4,000 translated products from 7,264 non-canonical open reading frames (ncORFs), termed peptideins. Functional validation and evolutionary analyses demonstrate their roles in cell essentiality and medulloblastoma survival, revealing a previously hidden class of biologically relevant peptides that expands the human proteome and offers new opportunities for peptide discovery and therapeutic development.

Intrinsically disordered regions regulate signaling and often drive disease, yet their flexibility and lack of stable pockets make them difficult to target. We developed deep-learning-based computational approaches to design de novo proteins that recognize flexible sequences, post-translationally modified epitopes, and aggregation-prone regions. These binders showed specificity and enabled applications in oncology, immunology, virology, and neurodegeneration, including CAR T targeting, MYC inhibition, HBx suppression, and Tau degradation. We also designed sequence-specific metalloproteases that cleave disease-relevant regions of TDP-43. Together, this work establishes a platform for converting dynamic molecular states into selective, therapeutically actionable targets across the proteome.

De novo peptide sequencing is an enabling technology for sequencing antibodies, neoantigens, non-canonical antigens, and non-ribosomal peptides, but existing methods remain error-prone. Recently introduced FFC-MS technology measures fragment–fragment correlations (FFCs), providing substantially richer information than conventional MS/MS spectra that record individual fragment ions. However, algorithms for peptide sequencing from FFC maps have been lacking. We present the first FFC-based peptide sequencing algorithm and demonstrate that it enables accurate reconstruction of long peptides, representing an important step toward reliable antibody sequencing.

We introduce a novel workflow for the discovery of protein-binding macrocyclic peptides from small libraries. Using this technology, we have identified a highly selective allosteric ligand for Protein Tyrosine Phosphatase 1B (PTP1B), an “undruggable” target in Type II diabetes. We also discuss how these ligands can be elaborated into protein degraders. Identification and characterization of a PTP1B allosteric ligand.

The development of targeted therapeutics requires balancing the exceptional specificity of antibodies with the manufacturability, tissue penetration, and chemical versatility of peptides. We have developed a peptide engineering platform that converts phenotypic screen-selected domain antibodies into structurally constrained cyclic peptides that preserve antibody-derived binding while enabling modular therapeutic conjugation. This presentation will describe our integrated discovery workflow, including phenotypic phage display, computational CDR3 analysis, cyclic peptide engineering, and functional validation. We will discuss how this platform enables the scalable development of compact, high-affinity targeting ligands for precision drug delivery and next-generation therapeutic applications.

The USP family of DUBs regulate Ub signaling using a common catalytic-domain fold. The dynamic nature of their catalytic-domain is important for controlling USP function, with inter- and intramolecular interactions influencing hydrolysis. This conformational flexibility, in combination with the high sequence conservation of the USP active site, has made it challenging to readily identify potent and selective inhibitors for individual USPs. Here, we demonstrate how a naive, macrocycle-mRNA display selection rapidly yielded high-affinity binders to USP7 that specifically inhibit the DUB in the nanomolar range. Structural analysis reveals distinct binding modes for several peptides, engaging hotspots previously identified by small-molecules.

Most chemical agents modulate membrane proteins by inducing conformational changes within their transmembrane (TM) helices, but act indirectly—binding at distal extra-membrane protein sites. We hypothesize that targeting TM domains directly within lipid provides a novel strategy for molecular modulation, yielding new tools and potential therapeutics. However, generating such chemical matter is difficult. We describe structure-based design strategies for peptides encoding molecular recognition of TM helices with biological functions.

Cyclic peptides have long been considered attractive as a drug modality due to their medium size and combining the advantages of small molecules and biologics. However, membrane permeability remains a significant challenge. Recently, physics-based methods combined with AI have emerged as a promising technology to design cyclic peptides with specific properties in mind. Here we focus on applying this method to design de novo cyclic peptides with drug-like oral bioavailability.