High-Throughput De Novo Protein Design Yields Novel Immunomodulatory Agonists 🚀 New preprint from David Baker!🚀 1. Researchers have developed a high-throughput de novo design approach to create novel cytokines, termed "Novokines," by fusing pairs of computationally designed binders targeting various receptor subunits. This method generated over a thousand potential Novokines, of which 75 activated pSTAT signaling in peripheral blood mononuclear cells (PBMCs). 2. The study identified new pairings of established common receptors, cross-family pairings such as TrkA-γcommon, and a series of pairings with interferon receptor-1 (IFNAR1), revealing that IFNAR1 can function as a versatile common receptor similar to γcommon or βcommon. 3. The framework provides a blueprint for expanding the understanding of cytokine signaling and generating novel therapeutic strategies. The designed binders are structurally programmable, allowing control over orientation and linker geometry, and can be produced at high throughput in bacteria. 4. The researchers characterized 18 Novokines, including those that drive monocyte proliferation, T cell survival, and CD4+ T cell-specific proliferation. The study also demonstrated that private receptors dictate signaling identity, while common receptors serve as modular scaffolds that enable diverse pairings. 5. The scale of the screen, the diversity of receptor pairings tested, and the use of primary human immune cells represent a substantial advance over previous efforts to engineer novel cytokine signaling. The findings suggest that the natural cytokine signaling network may reflect evolutionary constraints rather than absolute biochemical limits. 📜Paper: biorxiv.org/content/10.1101/… #ProteinDesign #CytokineSignaling #Novokines #SyntheticBiology #ComputationalBiology

Code to complex: AI-driven de novo binder design 1. This review by Fox et al. explores how AI has transformed de novo protein binder design, making it possible to rapidly generate high-affinity binders for diverse targets with improved efficiency and reduced resource requirements. The integration of AI into protein design marks a significant shift in the field, enabling the creation of custom binders that can neutralize toxins, modulate immune pathways, and engage disordered targets with high specificity. 2. The authors highlight the evolution of protein design from early rational design efforts to the current AI-driven approaches. They discuss how advancements in AI, such as the development of diffusion models and deep learning techniques, have enhanced the accuracy and flexibility of protein structure prediction and design. These improvements allow for the generation of binders with tailored architectures and functions, overcoming previous limitations in predicting sequence-structure relationships. 3. A key innovation is the use of generative diffusion models like RFdiffusion, which can design proteins with specific user-defined architectures and binding geometries. This approach, combined with sequence design models such as ProteinMPNN, has significantly increased the success rates of binder design compared to traditional methods. The integration of these tools into a powerful pipeline enables the in silico generation of binders with high experimental success rates. 4. The review also addresses the challenges and future directions in AI-driven protein design, including the need for better predictive accuracy, expanding the scope of targetable proteins, and addressing issues related to delivery and immunogenicity. Additionally, the authors emphasize the importance of ethical considerations, data integrity, and open access to computational resources to ensure equitable access and responsible use of these technologies. 5. The potential applications of AI-designed binders are vast, ranging from toxin neutralization and immune modulation to the development of novel diagnostics and therapeutics. Recent successes in designing binders for challenging targets like snake venom toxins and immune receptors demonstrate the real-world therapeutic potential of these approaches. As the field continues to advance, AI-driven protein design is poised to revolutionize medicine and biotechnology by enabling the rapid creation of highly specific and functional proteins. 📜Paper: cell.com/structure/fulltext/… #AIDrivenDesign #ProteinEngineering #DeNovoBinders #StructuralBiology #Biotechnology #TherapeuticDevelopment

Accurate de novo design of high-affinity protein-binding macrocycles using deep learning @nchembio 🚀 Paper published from David Baker!🚀 １．RFpeptides is a deep learning pipeline that enables de novo design of high-affinity macrocyclic peptide binders against diverse protein targets. It leverages a diffusion-based backbone generator (RFdiffusion) and a sequence designer (ProteinMPNN) with integrated DL and physics-based scoring. ２．The method succeeded in designing macrocycles that bind four distinct proteins (MCL1, MDM2, GABARAP, and RbtA) with medium to high affinity, testing only ~20 designs per target—orders of magnitude fewer than conventional display-based methods. ３．The standout binder, RBB_D10, targets a flat, previously uncharacterized site on Rhombotarget A (RbtA) with a dissociation constant (Kd) of 9.4 nM, despite starting from a predicted structure. The crystal structure of the complex closely matched the design model (Cα RMSD: 1.4 Å). ４．For GABARAP, two macrocycles (GAB_D8 and GAB_D23) exhibited Kd values of 6 nM and 36 nM, and IC50s of 0.7 nM and 2.5 nM in a competitive AlphaScreen assay—among the highest reported affinities for this target class. ５．Crystal structures for macrocycle–protein complexes (MCL1, GABARAP, RbtA) revealed close agreement with the design models, with RMSDs typically under 1.5 Å, validating the atomic accuracy of RFpeptides-generated binders. ６．Unlike traditional approaches, RFpeptides does not require known ligands or templates. It can target specific patches on proteins, including flat or non-canonical binding sites, by guiding the design via user-defined hotspots. ７．Binders adopt a range of structural topologies—α-helices for MCL1 and MDM2, β-sheets for GABARAP, and loop-like conformations for RbtA—demonstrating the structural diversity achievable with the pipeline. ８．Filtering of candidates combines DL-based (iPAE, pLDDT) and physics-based (ddG, SAP, CMS) metrics, followed by clustering and selection based on structural diversity, enabling rational downselection from tens of thousands of in silico models. ９．Even in the absence of explicit solubility constraints during design, successful binders exhibited good aqueous solubility, increasing the feasibility of downstream development. １０．RFpeptides achieves higher success rates and binding accuracy than any prior peptide design method tested, and allows structure-guided optimization without needing experimentally determined complex structures. １１．The approach opens the door to scalable, customizable design of macrocycles for therapeutics and diagnostics—potentially spanning intracellular targets, flat epitopes, or challenging pathogens with no structural data. 💻Code: doi.org/10.5281/zenodo.15264… 📜Paper: nature.com/articles/s41589-0… #ProteinDesign　#Macrocycles　#DeepLearning　#DeNovoDesign　#ComputationalBiology　#PeptideTherapeutics

Computational Design of Metallohydrolases 🚀 New preprint from David Baker!🚀 1. This study introduces RFam, a generative AI-based flow-matching model, which enables atomic-level scaffolding of enzyme active sites without predefined sequence positions or rotamer states, overcoming key limitations of previous methods like RFdiffusion. 2. Using density functional theory (DFT) descriptions of active site geometry, RFam designed 96 zinc metallohydrolases, achieving unprecedented catalytic efficiency, with the top design (A1) displaying a kcat/KM of 23,000 M⁻¹ s⁻¹—orders of magnitude higher than prior de novo designs. 3. The A1 enzyme features a novel fold with a highly preorganized active site and an enclosed pocket optimized for substrate positioning and catalytic efficiency, verified using the deep learning tool ChemNet. 4. RFam uniquely scaffolds functional groups rather than backbone motifs, significantly enhancing sampling of sequence space and enabling the generation of diverse and functional protein scaffolds. 5. Experimental validation showed 86 out of 96 RFam designs were soluble, and five demonstrated significant activity. A1 stood out with over 1,000 turnovers and remarkable stability during prolonged catalysis. 6. The study pioneers the integration of quantum chemistry, AI-driven design, and deep learning-based ranking to produce highly active, function-specific enzymes without the need for extensive directed evolution. 7. This approach opens doors to applications in environmental remediation and synthetic biology, as metallohydrolases can be tailored to degrade complex human-made pollutants efficiently. @samansalike31 @ikalvet @woodyahern 📜Paper: biorxiv.org/content/10.1101/… #EnzymeDesign #Metallohydrolases #ArtificialIntelligence #QuantumChemistry #SyntheticBiology #ProteinEngineering