However, because co-evolutionary methods do no construct a model of the relationship between individual sequences and structures, they are unable to predict structures for which no sequence homologs exist, as in new bacterial taxa or de novo protein design. A correct contact map can guide fragment assembly methods to an accurate 3D structure 25–50% of the time ( Ovchinnikov et al., 2017). With a large and diverse set of homologous sequences-typically tens to hundreds of thousands-co-evolution methods can accurately predict contact maps ( Juan et al., 2013). The second category of methods eschews explicit sequence-to-structure maps and instead identifies co-evolving residues within protein families to derive residue-residue contact maps, using co-evolution as an indicator of contact in physical space ( Hopf et al., 2014 Marks et al., 2011). Such template-based methods use one or more experimental structures-found through homology searches-as the basis for making predictions. Fragment assembly usually achieves high accuracy only when homologous protein structures are used as templates. This includes physics-based molecular dynamics simulations ( Marx and Hutter, 2012), which are restricted by computational cost to small proteins, and fragment assembly methods ( Gajda et al., 2011a), which find energy-minimizing conformations by sampling statistically-derived protein fragments. The first category builds explicit sequence-to-structure maps using computational procedures to transform raw amino acid sequences into 3D structures. Existing computational methods fall into two broad categories ( Gajda et al., 2011b, 2011a). Computational approaches to protein folding not only seek to make structure determination faster and less costly they aim to understand the folding process itself. Understanding how this occurs is a foundational problem in biochemistry. Proteins are linear polymers that fold into very specific and ordered three dimensional conformations based on their amino acid sequence ( Branden and Tooze, 1999 Dill, 1990). In the first task the model achieves state-of-the-art accuracy and in the second it comes within 1–2Å competing methods using co-evolution and experimental templates have been refined over many years and it is likely that the differentiable approach has substantial room for further improvement, with applications ranging from drug discovery to protein design. We test our model using two challenging tasks: predicting novel folds without co-evolutionary data and predicting known folds without structural templates. The model couples local and global protein structure via geometric units that optimize global geometry without violating local covalent chemistry. Here we introduce an end-to-end differentiable model for protein structure learning. Advances in deep learning that replace complex, human-designed pipelines with differentiable models optimized end-to-end suggest the potential benefits of similarly reformulating structure prediction. Co-evolution methods show promise, but an explicit sequence-to-structure map remains elusive. Predicting protein structure from sequence is a central challenge of biochemistry.
0 kommentar(er)
