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

Jonas Pfab; Dong Si

application range of homology modeling, since homologues proteins must be obtained first before making predictions, which is why we aim at the more challenging de novo structure prediction. Traditionally, research on de novo protein structure prediction was focused on inferring the 3D structure through the protein's amino acid sequence [8] . This approach is beneficial as the determination of the sequence is relatively straightforward. Methods that follow this approach either use geometric calculations to estimate the protein structure [9] , [10] , or they use deep learning techniques, such as Google's AlphaFold [11] . While these methodologies have led to impressive results, their accuracy is naturally limited as the amino acid sequence does not completely contain the protein's structural information, especially for very large protein complexes [12] .