protein design

GraphESMStable predicts protein mutation-induced thermal stability changes using a deep learning framework combining pre-trained sequence models and graph neural networks, achieving state-of-the-art performance.

This study identifies Artemis as a regulator of PRINS editing, showing that modulating Artemis improves insertion efficiency and specificity for genomic engineering.

This paper demonstrates intracellular bismuth-mediated cyclisation of peptides and miniproteins in E. coli, producing stable structures with minimal toxicity.

Contrastive scoring of protein variants across alternative backbone structures with inverse folding models (Conformational Biasing) predicts mutations that shift conformational equilibria and tune effector binding, catalysis, and labeling specificity.

The paper introduces MINT, a protein language model designed to model protein-protein interactions, outperforming existing models in tasks like binding affinity prediction and mutational effect estimation.

This paper introduces SidBait, a modular proximity ligation approach to identify small molecule and protein targets, validated with known interactions and applied to discover new targets like CaMKII and CapZ.

This paper designs soluble, miniaturized MHC proteins that enable stable expression in E. coli and retain peptide and TCR binding specificity.

This paper presents an AI-driven method to convert antibody sequences into functional intrabodies, successfully converting 19 out of 26 sequences, including those targeting histone modifications.

This paper designs de novo protein nanoparticles with integrated functional motifs, demonstrating their ability to form structured assemblies and elicit immune responses in mice.

This paper extends protein language models to a viral genomic scale using sparse attention, improving embeddings for long-range interprotein interactions.

This paper computationally redesigns T7 RNA polymerase to increase its thermal stability, achieving a T50 of 54.9°C with 59% wild-type activity at 37°C.

This paper designs novel acyl carrier protein variants using a computational method, showing they can undergo post-translational modification despite lacking the canonical helical structure.

RP3Net predicts recombinant protein production in E. coli using deep learning, achieving 77% accurate predictions and 92% correct identification of expressing constructs.

This paper designs metalloproteases that specifically cleave targeted peptide bonds in amyloid-β using a generative model, achieving high specificity and catalytic efficiency.

Disobind predicts inter-protein contact maps and interface residues for intrinsically disordered regions using sequence-based deep learning, outperforming existing methods without requiring structures or alignments.

BioGraphX improves subcellular localization prediction by using physicochemical graph encoding to provide interpretable and accurate results without requiring 3D structures.

This paper shows how evolution improves enzyme function by shifting the rate-limiting step from substrate binding to sub-state interconversion in β-lactamase OXA-48.

PocketX uses reinforcement learning to align protein pocket design with biophysical properties, outperforming existing methods on binding energy and evolutionary plausibility.

This paper presents the crystal structure of an engineered TdT variant and shows how mutations alter substrate specificity to accommodate bulky DNA terminators.

This paper introduces DREAM, a framework that enables protein design by inverting all-atom diffusion predictors, allowing gradient-based optimization of complex modalities like cyclic peptides.

This paper introduces OPENER, an engineered eRF1 variant that suppresses nonsense mutations and enables nonstandard amino acid incorporation, showing therapeutic benefits in a Duchenne muscular dystrophy model.

This study presents a joint optimization approach for modeling protein conformational landscapes and sequence design, showing improved stability prediction and sequence generation.

This paper engineers a protein that forms condensates in response to rare earth metals, enabling selective capture and release for sustainable recovery.

This study uses three protein language models to engineer TEV protease variants with improved stability and production, showing high efficiency with fewer designs.

SurfDiff predicts epitope selectivity and cross-reactivity by analyzing protein surface features, outperforming existing metrics in binder and immunogen design.

The paper develops a dual-inducer system to control gas vesicle expression in E. coli, preventing cell lysis and maintaining viability while enabling GV production.

This paper introduces a modular AAV retargeting platform using SpyTag/SpyCatcher for targeted gene delivery, demonstrating specific cancer cell transduction and therapeutic potential.

This paper designs globular proteins that drive biomolecular condensation and demonstrates their use in forming dynamic condensates in cells.

ProFam is an open-source protein family language model that improves fitness prediction and design by leveraging evolutionary context through homologous sequences.