Informace o projektu

Anotace je dostupná pouze v anglickém jazyce.

The ultrahigh-field NMR spectrometers with magnetic strengths of 1.2 GHz that cost more than 12.5 million Euros are within reach. The more powerful the magnet, the larger the protein structures to be revealed in general. The increases in sensitivity and resolution that arise from higher field strengths benefit the simplest type of experiments (NOESY) that report on interatomic distances and are used for NMR structure determination. The financial and technical investments required for these instruments call for methodological breakthroughs to exploit the better NOE build-up and sensitivity in larger proteins. Here, we propose to combine novel NOESY concepts with artificial intelligence to a streamlined strategy designed to meet the key objectives of NMR structure determination; minimal data collection, least human intervention, and applicability to large proteins. Artificial neural networks are to be trained using the massive amounts of chemical shifts deposited in Biological Magnetic Resonance Bank by human experts (8 million entries) and perform the NMR assignment task using only the 4D NOESY spectrum. Since Rosetta software can deliver accurate structures from a sparse network of distance constraints extracted from the NOESY spectrum, our approach promises to crack the atomic structures of large proteins in a single step. Today, image recognition by machines trained via machine learning in some scenarios is better than humans. The NMR assignment problem, if anything else, is a pattern recognition problem. As a major impulse for biomolecular NMR of large proteins, the present grant integrates the unique experience and knowledge in the analysis of NOESY data with several years of expertise in neural network applications to supersede approaches for protein structure determination in vogue that are highly time-consuming.