This presentation from September 2021 highlights the critical importance of characterizing the interactome—the vast network of molecular interactions—to understand biological function and the molecular basis of disease. It emphasizes the role of integrative modeling in bridging experimental gaps, using HADDOCK (High Ambiguity Driven DOCKing) as a primary example. By incorporating diverse experimental and bioinformatics data as Ambiguous Interaction Restraints (AIRs), HADDOCK drives a three-stage protocol involving rigid-body minimization, semi-flexible refinement, and solvent optimization to predict complex structures. It further highlights HADDOCK’s global impact, its application in COVID-19 research, and its future evolution into the modular HADDOCK3 framework for high-performance computing (HPC).

This tutorial uses the HADDOCK2.4 webserver and will demonstrate the use of HADDOCK for predicting the structure of a protein-protein complex from NMR chemical shift perturbation (CSP) data.

The best practice guide shows how to run HADDOCK in a sensible and rational manner. The guide will cover how to prepare structures for HADDOCK, how to use information about interactions in HADDOCK, how to analyse the results, which settings are best used in which scenario, and which are best avoided.  The guide will take you through all possible scenarios with related settings linked with tutorials of the newest HADDOCK version, published articles and protocols from the group of Prof. Alexandre Bonvin.

The ask.bioexcel HADDOCK category is meant for discussing any topic related to the use of the HADDOCK software, either as a local installation or via the HADDOCK web portal.

Link to a number of tutorials using the HADDOCK2.4 web server, ranging from basic to advanced, and including how to install HADDOCK2.4 locally.

Links to a number of basic tutorials on the usage of the new modular HADDOCK3 version.

This presentation from June 2021 highlights the importance of understanding the interactome—the complex 3D network of biomolecular interactions—to uncover the molecular basis of disease and advancing drug design. Since single experimental methods often fall short of providing complete structures, integrative modeling serves as a vital hybrid approach, combining high-resolution structural data with lower-resolution “puzzle pieces” from sources like mass spectrometry and bioinformatics. HADDOCK uses Ambiguous Interaction Restraints (AIRs) to actively drive the docking process rather than relying on a blind search. By employing a multi-stage protocol that includes rigid-body sampling, semi-flexible refinement, and solvent optimization, HADDOCK addresses the fundamental challenges of molecular flexibility and scoring, enabling the accurate atomistic characterization of diverse biomolecular assemblies.

This presentation from June 2021 explores the use of HADDOCK in the integrative modeling of diverse biomolecular complexes, emphasizing how experimental data can be used to drive—rather than simply score—structural predictions. Key case studies demonstrate HADDOCK’s effectiveness in modeling challenging antibody-antigen interactions by leveraging hypervariable loop data, as well as its ability to incorporate mass spectrometry (MS) and Cryo-EM data to resolve large macromolecular assemblies. Focus is placed on HADDOCK’s capacity to handle conformational flexibility and complex biological phenomena like fold-switching. By integrating automated fitting tools and modular refinement protocols, HADDOCK enables a shift from static structural models toward a dynamical representation of the interactome, providing a comprehensive framework for characterizing the functional landscape of molecular machines.

Two-part presentation on Computer Aided Drug Design hosted by Elixir

The second part of this presentation (starting at 37:40) from April 2023 marks the 20th anniversary of HADDOCK. Prof. Alexandre Bonvin discusses the evolution of the platform from its origins in NMR to its current capabilities in small molecule docking, specifically highlighting lessons learned from the D3R Grand Challenge. The presentation introduces a novel shape restraint protocol that bypasses the limitations of ‘apo’ receptor docking by utilizing 3D shapes and pharmacophore models derived from existing PDB templates. By employing ambiguous distance restraints and flexible refinement, HADDOCK achieves high-quality ligand poses even when significant conformational changes occur. Benchmarked against the DUD dataset, the protocol demonstrates a success rate of approximately 90% for top-10 models when a “safe” similarity threshold (Tanimoto coefficient >= 0.4) is met, proving to be a robust, competitive alternative for structure-based drug design.

This presentation from February 2026 explores the evolution of structural biology, tracing the shift from classical docking methods to the current AI-driven era led by AlphaFold. While AI has revolutionized protein structure prediction, significant challenges remain in modeling complex interactomes, particularly involving antibodies, RNA, and the persistent “scoring problem” of identifying accurate models from large ensembles. HADDOCK can address these challenges incorporating diverse experimental data to guide the docking process. By combining physics-based scoring with biochemical restraints, HADDOCK provides a robust framework for refining AI predictions and understanding the intricate molecular machines of life.

This presentation from February 2026 introduces HADDOCK3, a modular update of the traditional HADDOCK platform, designed to provide researchers with greater flexibility in integrative modeling workflows. While AI-driven tools like AlphaFold 3 have revolutionized protein structure prediction, this work highlights their limitations in modeling complex systems such as transmembrane ion channels and highly variable antibody H3 loops. By combining generative AI sampling (e.g., AlphaFlow) with physics-based refinement and experimental restraints, HADDOCK3 bridges the gap between static predictions and dynamic biological reality. Ultimately, the results underscore the vital role of integrative modeling in complementing AI to accurately resolve the intricate “social network” of the interactome.

First part of this tutorial which provides a practical guide to structural modeling of antibody-antigen complexes using HADDOCK3. The workflow emphasizes rigorous data preparation using PDB-tools, including chain merging, residue renumbering, and the removal of disordered regions or hetero atoms to ensure structural compatibility. Central to the approach is the use of Ambiguous Interaction Restraints (AIRs), which incorporate experimental or predicted data while accounting for uncertainty through a 50% random restraint deletion strategy. The simulation pipeline follows a structured sequence—from topology generation and rigid-body docking to flexible refinement and clustering—designed to yield high-resolution models of biological assemblies. The tutorial is optimized for versatile computing environments, including high-performance clusters and Google Colab, offering a scalable solution for characterizing protein-protein interactions.

Second part of this tutorial which provides a comprehensive overview of post-docking workflow analysis, focusing on the evaluation and validation of biomolecular complexes. By examining the modular output of the HADDOCK directory, users can track model rankings, interpret CAPRI evaluation metrics—such as the HADDOCK score, interface RMSD (iRMSD), and fraction of common contacts (fcc)—and utilize interactive HTML reports to visualize energy funnels and contact maps. The session highlights the importance of interface energetics, demonstrating how tools like alanine scanning and the PRODIGY-based “Proton” server can identify binding hotspots and predict the impact of mutations. Ultimately, the tutorial emphasizes that docking results must be critically validated against experimental data and structural energetics to move beyond simple model generation toward a deeper understanding of molecular recognition and interface stability.