Alexandre M. J. J. Bonvin

PowerFit improves computational performance for rigid-body fitting of atomic structures into cryo-EM maps while maintaining accuracy and providing interactive visualization.

MIMOSA improves de novo antibody design by constraining generative models to known binder interfaces, achieving validated binders on targets where unconstrained methods fail.

The paper characterizes two anti-HPA-1a antibodies from a severe FNAIT case, showing functional differences in platelet activation through FcγRIIa engagement.

AbTune improves protein language model performance for antibody-related tasks through depth-selective fine-tuning, outperforming baselines in structure prediction and mutation effect prediction.

DeepRank-Ab improves scoring of antibody-antigen complexes using geometric deep learning, outperforming existing methods in accuracy and ranking.

The paper uses AlphaFold to generate full-length structural models of mitofusins, revealing a novel cross-type dimerization mode and a new hypothetical mechanism of outer mitochondrial membrane fusion.

The paper integrates the MARTINI2 force field into HADDOCK3 to enable faster modelling of large biomolecular complexes with reduced computational costs while maintaining accuracy.

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

HADDOCK3 is a new modular platform for integrative biomolecular complex modelling, offering enhanced flexibility and integration with third-party tools to complement machine learning predictions.

This study identifies the mumps virus SH protein as a pentameric viroporin and shows it functions as a chloride channel inhibited by BIT225.

The paper improves antibody structure prediction by combining AlphaFlow and HADDOCK to generate diverse CDR-H3 loop conformations, enhancing docking accuracy for antibody-antigen complexes.

This paper introduces a method to generate bound-like conformations of flexible, multi-domain proteins using unbound structures and binding site information, demonstrating its effectiveness on adenylate kinase.

The paper summarizes HADDOCK's performance in CAPRI rounds 47-55, highlighting the effectiveness of integrating HADDOCK with AI-driven methods like AlphaFold2 for protein complex structure prediction.

The paper evaluates how machine learning-generated antibody and antigen structures can improve antibody-antigen complex modelling using HADDOCK, showing improved accuracy over ZDOCK.

This study demonstrates the use of HADDOCK for predicting protein-glycan complexes, achieving 70% and 40% top 5 success rates on bound and unbound datasets, respectively.

ARCTIC-3D retrieves and clusters protein-protein interface data from structural information to classify interfaces and model biomolecular assemblies.

The paper evaluates reducing AlphaFold-Multimer model sampling for faster protein-protein interaction predictions without compromising accuracy.

This paper introduces DeepRank-GNN-esm, a graph neural network that uses ESM-2 embeddings to score protein-protein models, replacing computationally expensive PSSM features and improving performance on docking and crystal artifact detection tasks.

The paper applies DisVis to filter co-evolution predicted contacts for protein complexes, showing it improves low-quality data but HADDOCK remains robust to false positives.

MetaScore improves traditional protein-protein docking scoring functions using machine learning, outperforming existing methods in success and hit rates.

DeepRank-GNN uses graph neural networks to score protein-protein docking models, outperforming existing methods and improving speed and storage efficiency.

The paper presents a refinement method using HADDOCK2.4 to reduce atomic clashes in low-to-medium resolution cryo-EM complexes by optimizing protein-protein interfaces.

This paper presents a protocol for modeling cyclic peptides and their interactions with proteins using HADDOCK2.4, achieving 100% success rate in predicting acceptable models for a dataset of 30 complexes.

DeepRank uses 3D CNNs to mine protein-protein interfaces, achieving competitive or better performance in classification and docking model ranking.

The paper evaluates using molecular dynamics simulations with machine learning to distinguish native from non-native protein-protein docking models, achieving 0.85 accuracy with simulations of 50-100 ns.

The paper introduces two new HADDOCK protocols for protein-small molecule docking using shape restraints from homologous compounds, achieving an 81% success rate on unbound docking benchmarks.

This paper introduces a membrane-driven docking protocol for modeling membrane-associated protein complexes using AI-based methods and evaluates its performance.

proABC-2 predicts antibody-antigen contacts using a convolutional neural network and integrates with HADDOCK for docking.

The paper introduces iScore, a software tool that ranks protein-protein docking models using a random walk graph kernel and support vector machines, with MPI and GPU support for efficient computation.

The paper evaluates HADDOCK's performance in CAPRI rounds 38-45, showing it produces acceptable or better models for most targets using data-driven strategies.