DeepTracer: Predicting Backbone Atomic Structure from High Resolution Cryo-EM Density Maps of Protein Complexes

Jonas Pfab; Dong Si

In order to evaluate the runtime of the DeepTracer for larger proteins we predicted the EMD-4903 density map (echovirus 1 intact particle). The number of predicted atoms for this map was 32,446. The execution of the end-to-end prediction pipeline of the DeepTracer took only two hours. As a point of comparison, the expected runtime of the C-CNN method for the same density map is more than 5 days assuming the trend in In addition to the location and secondary structure of amino acids, the DeepTracer also predicts their types. As reported in Table 1 we assigned the correct type out of the 20 possible amino acid types with an accuracy of 25.2%. In order to get a better idea of what that result means we look at a specific structure prediction. For the EMD-6272 density map, we predicted the correct amino acid type for 31.49% of amino acids. Note that this is predicted solely from 3D map without using any genetic information of the protein. In Fig 12 The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.02.12.946772 doi: bioRxiv preprint solved structure. The sequence was created by walking along the predicted chain of amino acids and noting down the type of each amino acid. The predicted amino acids from 3D could potentially be utilized for the reference of future full-atom structure prediction.