Methods of computer modelling of the transcription factor of stress resistance WRKY2 in Triticum Aestivum: A comparative analysis

The aim of this work was to compare the efficiency of modern computer modelling methods for constructing the three-dimensional structure of the transcription factor TaWRKY2 from Triticum aestivum, which plays a key role in plant stress resistance. The study used traditional homology approaches and machine learning-based methods, including AlphaFold, RoseTTaFold, ESMFold, and OmegaFold. The model obtained with AlphaFold2 (MMseqs2) was the best in terms of Clashscore (99th percentile), MolProbity Score, and ERRAT. The complex with zinc ions and DNA generated by AlphaFold3 showed better results in terms of Z-Score. The AlphaFold method provided biologically relevant structures with the correct location of functional sites. ESMFold provided fast modelling, but showed deviations in the structure, including additional α-helices and lower quality in most metrics. The RoseTTaFold model had an elongated shape with a higher content of α-helices and required additional verification of functional activity. OmegaFold, although it provided the best QMEAN score (0.4), was inferior in other metrics. Additional tools, such as Amber and Chimera, allowed for structure relaxation and analysis of key features, including the spatial arrangement of the zinc fingers (18.7 Å). Evaluation by Verify 3D, ERRAT, Ramachandran Plot, and other metrics revealed the advantages and disadvantages of each approach. The findings confirmed the advantages of machine learning methods for modelling proteins with high functional plasticity. In particular, AlphaFold was recognised as the most effective approach for building models with high accuracy. At the same time, the use of several methods allows considering alternative conformations and interactions, which is important for a deeper understanding of the functional mechanisms of the protein