hide
Free keywords:
-
Abstract:
Deep learning-based tools revolutionized the field of protein structure prediction. We present our recent attempts to use one of such methods, AlphaFold2 (AF2), for the modeling of coiled-coil domains. We benchmarked AF2 using a set of experimental coiled-coil structures. For each benchmark case, we applied the following procedure: first, we determined local structural parameters [1] for the experimental structure and the corresponding AF2 model. Then, by comparing these structural parameters, we defined a quality index that reflects the discrepancy between the experimental structure and the model. Knowing the advantages and limitations of AF2, we performed more focused analyses of (i) HAMP domains [2], a family of coiled coils whose conformational changes are key for prokaryotic signal-transducing proteins, and (ii) the applicability of AF2 models for the augmentation of data sets used for training of machine learning methods.
