protein structure

The paper evaluates AlphaFold's predictions of transmembrane protein complexes, finding that while most show plausible membrane geometry, some have inconsistencies and limitations in predicting lipid interactions.

The paper introduces WhatRes, a tool to evaluate AI-predicted protein structures by comparing their geometry to experimental structures across resolution and realism.

This study evaluates AlphaFold 3's accuracy and reliability in predicting RNA structures and protein-RNA interfaces, finding that confidence scores do not always correlate with correct predictions.

Struct-Caliper improves protein structure search reliability by providing accurate p-values, increasing homolog recovery and reducing false discoveries compared to native E-values.

The paper uses computational methods to design and evaluate splicease variants for generating diverse ketoamide substrates, finding that top-scoring peptides are spliced more efficiently than single-point mutants.

EpiRank evaluates epitope prediction by ranking surface patches on single-chain geometry, finding it correctly identifies real epitopes 64% of the time but with significant baseline overlap.

This study compares docking and short MD rescoring for ranking p38α inhibitors and finds no significant improvement from MD rescoring over docking.

The paper benchmarks structure prediction models on a new antibody dataset to assess their ability to distinguish binders from decoys, finding moderate success but limitations in epitope localization.

The paper evaluates multimodal protein foundation models for allosteric binding site detection, finding that dataset properties, not model complexity, primarily determine performance.

This study shows human macrophages have functional CYP3A4 activity, with expression and function discordance, and cocaine increases CYP3A4 activity across donors.

This study benchmarks AlphaFold 3's interaction scores against BANDOCK+ for drug discovery, showing AF3's slightly better performance in disease ranking but BANDOCK+'s broader coverage.

The paper introduces MutaStruct, a curated benchmark for evaluating mutation-induced structural changes, showing that overall structural fidelity does not guarantee accurate modeling of mutation effects.

The paper introduces DEEP-Boltz, a framework that extracts energetic potentials from protein structure predictors, outperforming existing methods in predicting mutation-induced stability changes.

Chaperone, a DataFrame-based framework and agent harness for protein design tools, cut tokens 86% and GPU-hours 64% while raising pass rate from 6% to 12%.

MIEN improves glycan pose ranking by distinguishing experimentally observed conformations from non-native ones in protein-carbohydrate interactions.

This study uses machine learning to predict bNAb-HIV-1 Env binding energies across clades, revealing clade-specific binding patterns and affinity trends.

The paper presents a manually curated dataset of protein binding domains, linking them to molecular interactions and genomic variations for benchmarking and disease mechanism studies.

The paper investigates how structural templates affect apparent links between protein stability predictions and longevity in tetrapods, finding that template choice can create spurious associations.

This paper introduces a computational pipeline for de novo metalloenzyme design using Metal Mer, achieving high selectivity and activity in catalytic reactions without directed evolution.

The paper benchmarks protein interaction models to distinguish direct partners from non-binding or noncontacting proteins, showing that context recognition and direct specificity are separable capabilities.

The paper reports that two BindCraft2 campaigns produced no binders despite high interface confidence, identifying issues with confidence metrics and a disabled feature.

On 489 evaluable new Mac1 complexes, AlphaFold3, Chai-1 and Boltz-2 placed over half of ligands within 2 Å, but on docking hit-lists AF3 and Boltz-2 added little over docking.

This study evaluates LLMs' ability to predict mutation effects based on structural data without using dedicated predictors, finding comparable accuracy to existing methods.

AlphaFold3 pLDDT matches AlphaFold2 at calling disordered residues (Matthews correlation 0.693 vs 0.701); on disorder-containing complexes mean DockQ is similar, but interface features explain less variance (42% vs 58%).

The paper accelerates structural analysis in HADDOCK3 by reformulating RMSD and FCC clustering as GPU-based linear algebra operations, achieving significant speedups without compromising accuracy.

The paper benchmarks interface confidence as a selection criterion for computational nanobody design against survivin's BIR-domain groove and finds it is not correlated with epitope engagement.

This paper introduces autofocusing, a method that improves molecular simulations by integrating experimental data, demonstrating its effectiveness for phospholipid bilayers.

This paper shows that deep learning models' performance in predicting protein-ligand binding poses depends on the energy landscape's frustration quenching and network topology, not algorithmic flaws.

The paper identifies a sunflower FLS2 receptor that recognizes a flg22 escape variant and uses structural and engineering approaches to reconstruct recognition in divergent FLS2 homologues.

The paper introduces RNA3D-Decoy, a benchmark dataset of AI-generated RNA structure models, to evaluate quality assessment methods for RNA structure prediction.