Protein design articles from across Nature Portfolio

The discovery of macrocyclic peptide therapeutics has been slow. We introduce RFpeptides, a deep learning method that enables the de novo design of macrocyclic peptide binders to therapeutic targets. The designed macrocycles bind their respective protein targets with high affinity and atomic-level accuracy.

Deep learning-based generative tools are used to design protein building blocks with well-defined directional protein bonding interactions, allowing for the generation of a variety of scalable protein assemblies from a small set of reusable subunits.

Designing and optimizing proteins by mutagenesis suffers from the overwhelming space of possible variants. A recent study developed µProtein, a reinforcement learning model coupled with a protein language model as a surrogate oracle, to accelerate this process towards high-functioning proteins.

De novo design of Ca²⁺ channels with ion selectivity that can assemble appropriately and mediate Ca²⁺ conductance when expressed in cells is described.

The authors introduce a method that combines physics and machine learning to design dynamic unstructured proteins with tunable ensemble properties like size, shape, sensing and binding.

This study presents FolTAC-dual, a folate receptor-mediated platform for dual degradation of EGFR/HER2 and PD-L1/VISTA, offering a strategy to overcome drug resistance and enhance antitumor immunity for cancer treatment.

Jian Ma et al. present HuDiff, a diffusion-based deep learning framework that humanizes antibodies and nanobodies (a small type of antibody) without templates. The model achieves improved humanness while preserving or enhancing binding strength, and the authors show promising results in virus neutralization experiments.

Coagulation factor IX (FIX) contributes to hemostasis through both plasma activity and binding to extravascular collagen IV. Here, the authors show that collagen binding of albumin-fused hyperactive FIX can be engineered to tailor the pharmacokinetics, distribution, and functional properties in mice for tailored short- or long-term hemophilia B therapy.

The discovery of macrocyclic peptide therapeutics has been slow. We introduce RFpeptides, a deep learning method that enables the de novo design of macrocyclic peptide binders to therapeutic targets. The designed macrocycles bind their respective protein targets with high affinity and atomic-level accuracy.

Deep learning-based generative tools are used to design protein building blocks with well-defined directional protein bonding interactions, allowing for the generation of a variety of scalable protein assemblies from a small set of reusable subunits.

Designing and optimizing proteins by mutagenesis suffers from the overwhelming space of possible variants. A recent study developed µProtein, a reinforcement learning model coupled with a protein language model as a surrogate oracle, to accelerate this process towards high-functioning proteins.

PepMLM, a protein language model fine-tuned on protein–peptide data, can generate potent, target-specific linear peptides capable of binding to and degrading proteins ranging from cancer receptors to drivers of neurodegeneration and viral proteins, all without requiring protein structural information.

ProDomino is a machine-learning model that efficiently predicts domain insertion sites in host proteins on the basis of amino acid sequence alone. The model enables the greatly accelerated design of functional multi-domain proteins, such as light-triggered or drug-triggered protein switches.

Self-labeling protein tags (SLPs) have revolutionized bioimaging by enabling precise protein tracking. A next-generation SLP, SNAP-tag2, has now been developed with improved labeling kinetics, fluorescence brightness, and substrate accessibility. These enhancements enable deeper insights into cellular processes in the field of bioimaging.