AlphaFold2 – Machine learning revolution in structural biology


Karel Berka, Palacký University Olomouc, ELIXIR CZ
Marian Novotný, Charles University, ELIXIR CZ

An algorithm AlphaFold was called a scientific breakthrough of the year 2021 by the Science journal – the AlphaFold algorithm predicted the 3D structures of proteins from the sequence better than its competitors in the CATH competition in quality indistinguishable from the experimental structure. Since late 2020, AlphaFold has been used to predict 3D structures in a variety of scenarios, including those for which it has not been trained at all. A database of AlphaFold models was created for common use in July 2021, and other tools have been adopted to a new reality of having a powerful tool to predict 3D structure from literally any protein sequence. We are slowly beginning to answer the following questions: How does AlphaFold actually work? How difficult is it to run AlphaFold with my data? How to interpret the AlphaFold models? What tasks are still difficult for AlphaFold? Would we still need an experimental determination of protein structure? In our lecture, we will try to address these questions as much as possible, based on our practical experience with protein structure models.

The lecture will take place online.