Thumbnails:
List:
Year:
Category:
Session:
Poster:
Getting poster data...
Mark Rozanov. Haim J. Wolfson (Tel Aviv University)
The cryo-EM resolution revolution enables the development of algorithms for direct de-novo modelling of protein structures from given cryo-EM density maps. Deep Learning tools have been applied to locate structure patterns , such as rotamers, secondary structures and C-alpha atoms. We present a deep neural network for the semantic segmentation of a cryo-EM density map. The developed network labels voxels in a cryo-EM map by the amino acid type of the sampled protein structure A group convolution network combined with UNET was used for semantic segmentation of a high resolution cryo-EM map. The rotation invariant property of group CNNs enables achieving high accuracy even for a relatively small training dataset. As a map resolution degrades the ability to identify amino acid types diminishes, imposing an upper bound on a detection algorithm performance. However, from the structural biology point of view, only the correctly detected amino acids are of interest. Consequently, an additional group CNN was trained to distinguish mislabelled voxels from the correctly labelled ones. The combination of the two above NNs reports only voxels which have been labelled with high