AI-based integrative structural modeling

Jan Kosinski

EMBL Hamburg [DE]

Macromolecular assemblies are essential to biological processes, yet their structural characterization remains challenging due to their size and complexity. Integrative structural modeling addresses this by combining data from various experimental techniques, including electron microscopy, X-ray crystallography, and crosslinking mass spectrometry, with computational approaches. In this presentation, I will discuss our recent advancements in integrative modeling of macromolecular assemblies, emphasizing the incorporation of AI-based structure prediction tools such as AlphaFold. We have utilized these tools to construct a detailed structural model of the human nuclear pore complex (NPC). The NPC, a 120-megadalton assembly regulating nucleocytoplasmic transport, has posed significant challenges for structure determination due to its size and flexibility. By integrating cryo-electron tomography data with AlphaFold predictions, we achieved a comprehensive model covering approximately 90% of the structured scaffold. Additionally, I will present AlphaPulldown, a Python package designed for high-throughput modeling of protein–protein interactions using AlphaFold-Multimer. This tool streamlines the screening process and facilitates the modeling of higher-order oligomers. Furthermore, I will present AF3x, a method that enhances structural modeling by incorporating crosslinking mass spectrometry restraints in explicit atomic representation to drive AlphaFold 3 predictions. This approach can sometimes improve the accuracy of AlphaFold 3 models, particularly for complexes with limited evolutionary data.

Go back