Protein Structure and Dynamics

The paper demonstrates a quantum-inspired kernel achieves high accuracy and Matthews correlation coefficient in classifying protein torsion angles, matching classical models.

The paper uses deep learning to design novel protein kinase sequences with multiple mutations, measures their activity, and identifies key sequence determinants of function.

This study maps aggregation-prone regions in the myocilin OLF domain using molecular dynamics simulations to identify potential therapeutic targets for POAG.

The paper characterizes the multi-timescale structural dynamics of the p53 transactivation domain using NMR and simulations, revealing a hierarchical energy landscape.

NEXT-FRET identifies protein folding intermediates under native conditions using a solution-based smFRET platform with a time-variant Gaussian Mixture Model.

The paper predicts pH-dependent sites in the human proteome, finding enrichment in cell adhesion proteins and shielding of ribosomal/nuclear processes from pH fluctuations.

This study evaluates how well AlphaFold2, ESMFold, and ProteinMPNN predict the success of de novo protein designs using confidence metrics, finding that ESMFold performs best but overall accuracy is limited.

GeoGAT-site predicts protein-protein interaction sites using a geometric graph attention network, achieving high accuracy and efficiency compared to existing methods.

This paper evaluates AI methods for predicting antibody-antigen complexes, introduces AntiConf as a confidence metric, and shows it improves precision and recall for these predictions.

This study uses atomistic simulations to show that Lanmodulin can fold into a native-like structure without metal ions, challenging the belief that metal binding is required for its folding.

This study quantifies how domain extensions in a PDZ protein reshape its energy and allosteric landscapes through extensive mutational analysis.

The paper introduces the Dayhoff Atlas, a resource with large protein sequence datasets and generative models to improve protein design and prediction.

This paper introduces modality-chain reasoning for multimodal protein modeling and design, showing improved performance in structure prediction, inverse folding, and fitness prediction compared to existing models.

The paper develops containerized solutions and an optimized platform for protein structure prediction and design tools to improve accessibility and computational efficiency.

The paper introduces PXMeter, a toolkit for evaluating structure prediction models, and benchmarks several models on a curated dataset from the PDB.

The paper shows that structure alignment algorithms overestimate significance and introduces a new method for accurate statistical significance estimation.

The paper introduces Auto-MartiniM3, a tool for automating the coarse-graining of small molecules for Martini 3 simulations, validated on 81 molecules and complex cases like caffeine.

AntiDIF improves antibody inverse folding by increasing sequence diversity while maintaining high sequence recovery rates and structural accuracy.

ProVADA adapts proteins to non-native contexts by steering a pretrained generator while preserving function, successfully converting renin and deimmunizing antibodies.

AlphaFlex improves prediction of multiple protein conformations by integrating flexible residue prediction and directed masking, outperforming existing methods on apo-holo pairs.

The paper compares LigandMPNN and directed evolution for redesigning the effector-binding site of the RamR transcription factor, finding limited overlap between the two methods' results.

This study investigates how counterion valency affects the electrostatic collapse of an intrinsically disordered protein, showing divalent cations induce greater compaction than monovalent cations.

The paper benchmarks AlphaFold, ESMFold, and Boltz in maize to assess their performance on plant proteins, finding lower confidence scores and geometry issues in less conserved proteins.

Treating low-confidence AlphaFold structures as noisy data at pLDDT-dependent diffusion times lets a 16.7M-parameter model generate long proteins with higher diversity, designability and novelty than Proteína.

The paper investigates how missense mutations in Shank1 PDZ domain affect ligand-specific binding affinity through dynamic interactions.

The paper uses Boltz-2 to predict structural models of human protein interactions and identifies functional and cancer-related networks.

TDFold improves single-sequence protein structure prediction using a 2D geometric template diffusion method, outperforming existing models on homology-insufficient datasets and offering lower resource consumption and faster inference.

This paper introduces FoldCraft, a fold-conditioned AlphaFold2-Multimer hallucination framework for de novo binder design, achieving higher structural and binding confidence than existing methods.

This study evaluates machine learning methods for nanobody structure prediction and benchmarks HADDOCK workflows to model nanobody-antigen interactions using AlphaFold2 and ImmuneBuilder predictions.

The paper develops a computational method to design proteins that bind small molecules tightly and specifically, validated experimentally with X-ray crystallography and functional assays.