3D structures of biomolecules – “dictionaries”

A German and US research team led by Heinrich Heine University (HHU) Düsseldorf has developed a data description that can provide fluorescence measurements for structural and dynamic modeling of large biomolecules. The authors explain in a scientific journal Natural Methods that for the first time, other researchers can access fluorescence-based integrative structural models and their dynamics through databases. This provides experimental training data for the next generation of artificial intelligence tools for modeling dynamic structures.

Proteins and nucleic acids are the central building blocks of life in all organisms. These biomolecules are made up of many individual building blocks, such as amino acids in the case of proteins. When individual building blocks are assembled in cells, biomolecules form as complex three-dimensional structures. Their specific shape is determined by the configuration and forces between the building blocks. However, to understand the function of biomolecules, it is important to take into account not only their three-dimensional structure, but also the number of different structural states and the dynamics of exchange between them.

For a long time, determining the three-dimensional structure of biomolecules using classical biophysical methods was very difficult and time-consuming. To simplify and organize this work step by step, all these 3D structures have been collected in the Protein Data Bank (PDB) since 1971 (link: https://www.rcsb.org). These more than 220,000 structures are used by artificial intelligence tools such as AlphaFold, which won this year’s Nobel Prize in Chemistry, as training data for neural networks. Artificial intelligence systems are already making good predictions about biomolecular structures. However, instruments are currently unable to predict dynamics due to a lack of experimental data.

It is therefore important to use powerful experimental techniques such as fluorescence spectroscopy, which provide comprehensive information about the dynamics and structure of complex biomolecules. In fluorescence experiments, certain interesting parts of biomolecules are marked by small dye molecules that light up when externally excited and thus reveal their position. Integrative modeling approaches combine such experimental data with computational methods to achieve higher structural resolution and account for dynamics.

Dr. Christian Hanke, postdoc at the HHU Institute of Molecular Physical Chemistry and first author of the paper published in Nature Methods, emphasizes: “Fluorescence experiments provide detailed information, making them an excellent source of data for integrative modeling. However, to take full advantage of this wealth of information, it must be accessible and usable by the wider scientific community.”

In the publication, a research team from HHU, Rutgers State University of New Jersey, and the University of California, San Francisco presents a standardized description of data in the form of three linked “vocabularies” that are organized into a shared library. Prof. Dr. Klaus Seidel from HHU, one of the two corresponding authors: “This organizational principle with combined vocabularies allows researchers to store integrative structural models based on fluorescence data together with kinetic information for the first time. At the same time, the general definitions can be used by other methods to document the dynamic properties of biomolecules along with their structure in a database.”

This approach is necessary to relate static structural models to their underlying energy landscape, that is, the energetic differences between different three-dimensional arrangements of building blocks within a biomolecule. Professor Seidel: “This information allows us to develop and train a new generation of artificial intelligence programs to predict dynamic biomolecular structures. This is where data obtained from fluorescence experiments on functionally relevant dynamics can make a very important contribution.”

This work was carried out under the ERC extended grant “hybridFRET” to Professor Seidel.

Original publication:

Christian A. Hanke, John D. Westbrook, Benjamin M. Webb, Thomas-O. Peulen, Katherine L. Lawson, Andrei Sali, Helen M. Berman, Klaus A. M. Seidel, and Brinda Vallat. Ensuring accessibility of integrative structures based on fluorescence and associated kinetic information. Natural Methods (2024).

DOI: 10.1038/s41592-024-02428-x