Methods to accelerate protein structure determination by solution NMR

核磁共振化学位移为蛋白质提供了重要的局部结构信息。从核磁共振化学位移数据生成一致的结构对于大约100-130个残基的蛋白质来说是可行的，并且这种结构的质量与用标准核磁共振协议获得的结构相当。在与David Baker博士和他的团队的合作中，我们开发了一种化学位移引导的方法，以化学位移为基础成功准确地确定结构，但在实践中，这种方法仅限于相对较小的蛋白质。新的工作重点是扩展这种方法，以允许合并容易获得的实验信息，并更广泛地利用先前解决的结构的可用数据库。

NMR chemical shifts provide important local structural information for proteins. Consistent structure generation from NMR chemical shift data has become feasible for proteins with sizes of up to about 100-130 residues, and such structures are of a quality comparable to those obtained with the standard NMR protocol. In collaboration with Dr. David Baker and his group, we developed a chemical-shift-guided approach to successfully and accurately determine structures on the basis of chemical shifts, but in practice the approach was limited to relatively small proteins. New work focuses on extending this approach to allow incorporation of easily accessible experimental information and more extensively exploit the available database of previously solved structures. By means of an optimized neural network algorithm, SPARTA+, we are able to estimate chemical shifts for proteins of known structure. Using this program, we are able to assign approximate chemical shift values to crystallographically determined structures. Standard bioinformatics tools, that have been designed originally for searching for proteins of remotely homologous sequence, can be adapted to now search the database for sequences with homologous chemical shifts and thereby find protein chains of similar structure, without requiring any sequence homology. The approach is proving to be very robust, and is able to use existing algorithms to deal with gaps in the sequence when searching for structural homologs. The method increases in efficiency with the size of the protein and therefore represents an ideal complement to the CS-Rosetta approach developed earlier by us in collaboration with the Baker group. It does require, however, the presence of prior solved protein of a similar fold in the database and will not reach convergence if no good structural templates can be found. The approach is computationally rather demanding, and requires access to cluster computing.