Date: 8 April 2025

Time: 15:00 CET

Registration

Abstract

Antibodies are specialized proteins used by the immune system to eliminate unrecognized, potentially harmful molecules (antigens). Their ability to bind antigens with high specificity makes them an ideal molecule to work with in pharmaceutical research and drug development. However modelling antibodies offers endless challenges from a structural perspective, which have been only partially addressed by the recent developments in machine learning-based structure prediction (e.g. AlphaFold21, AlphaFold32). These algorithms tend to rely on coevolutionary information, which is missing in both the antibody’s Complementarity-Determining Regions (CDRs) and between the antibody and antigen sequences.

In this talk I will discuss when and how it is possible to obtain accurate structural predictions of these proteins and their complexes. I will demonstrate how integrating experimental data with AI-driven modeling within the BioExcel flagship software HADDOCK3-4  improves prediction accuracy5. Such models can then be used as a starting point for improving the binding properties of the complexes through antibody design.

I will showcase real-world examples of antibody structural prediction challenges, focusing on cases where pure machine learning-based prediction is unsuccessful.

 

References

  1. Highly accurate protein structure prediction with AlphaFold J Jumper et al Nature 596 (7873), 583-589
  2. Accurate structure prediction of biomolecular interactions with AlphaFold 3. J Abramson et al Nature 630.8016 (2024): 493-500
  3. HADDOCK: a protein-protein docking approach based on biochemical or biophysical information C Dominguez, R Boelens, AMJJ Bonvin Journal of the American Chemical Society 125 (7), 1731-1737
  4. The HADDOCK2.4 Web Server: A Leap Forward in Integrative Modelling of Biomolecular Complexes. Honorato, R. V. et al. Nature protocols 19 (11), 3219-3241
  5. Towards the accurate modelling of antibody-antigen complexes from sequence using machine learning and information-driven docking M Giulini et al Bioinformatics 40 (10), btae583

Presenter

Marco Giulini (he/him), Bijvoet Centre for Biomolecular Research, Faculty of Science – Chemistry, Utrecht University

2012-2015: Bsc in Physics, University of Milano

2016-2018: Msc in Quantitative and Computational Biology, University of Trento

2018-2022: PhD in Physics at The University of Trento with Prof. Raffaello Potestio – focusing on method development for coarse-grained modelling of biological systems.

2022: present: Postdoc in Prof. Alexandre Bonvin’s group at Utrecht University – working on computational structural biology.

Marco’s research is focused on combining physics-based approaches and machine learning algorithms to predict the structures and properties of biomolecular systems. He is primarily involved in developing the new BioExcel version of the HADDOCK docking tool (HADDOCK3), and improving antibody and antibody-antigen structure prediction methods.