Descrizione del progetto
Struttura proteica ad altissima risoluzione sulla base di misurazioni RMN
La spettroscopia a risonanza magnetica nucleare (RMN) è essenziale per l’analisi della struttura proteica in quanto permette di misurare la dinamica e la struttura di una proteina in condizioni quasi fisiologiche. Recenti studi sull’effetto nucleare Overhauser (NOE, Nuclear Overhauser Effect) esatto hanno permesso di misurare la distanza nelle proteine tramite RMN con una precisione di 0,1 Å. Il progetto DeepNOE, finanziato dall’UE, si propone di sviluppare un innovativo modello/algoritmo che trasforma automaticamente le misurazioni RMN grezze in strutture proteiche ad alta risoluzione che rivelano in dettaglio atomico più stati conformazionali popolati simultaneamente. Lo studio comprenderà l’applicazione dell’apprendimento profondo, un campo nuovo nell’apprendimento automatico che ha rivoluzionato la scienza dei dati e l’intelligenza artificiale.
Obiettivo
Nuclear Magnetic Resonance (NMR) spectroscopy is one of the leading techniques for protein structure analysis. In contrast to other methods, NMR spectroscopy allows the measurement of the dynamics and structure of a protein under nearly physiological conditions, without the need for crystallization or freezing of a sample. Recent studies on exact Nuclear Overhauser enhancements (eNOEs), carried out in the laboratory of the host professor, have enabled distance measurements in proteins by NMR with accuracy of 0.1 Å. This allows to determine structures in solution and in living cells with unprecedented resolution.
This biophysical achievement creates an outstanding opportunity for a computer scientist (the fellow candidate) to develop the first-of-its-kind model/algorithm that automatically transforms raw NMR measurements into high-resolution protein structures that reveal multiple simultaneously populated conformational states in atomic detail. This problem will be tackled with the use of deep learning (DL), a novel field in machine learning that has emerged after 2010 and has revolutionized data science and artificial intelligence.
The project is divided into 3 parts. First, it is planned to investigate recent advances in DL to derive a model that extracts visual information from 2D and 3D NMR spectra. Afterwards, the proposed model will be integrated into CYANA to formulate a hierarchical DL/optimization routine, which automates all steps of protein structure solving. Finally, it is planned to explore the possibility of calculating protein structures directly from NOESY spectra, which constitutes a new protocol for protein structure solving by NMR spectroscopy.
Summing up, the proposed DL approach has the potential to reduce the time required to solve proteins with NMR from months/years to days, while delivering very high resolution, multi-state structures. We expect this project to open new avenues in structural biology and drug discovery.
Campo scientifico
- natural sciencescomputer and information sciencesdata science
- natural sciencesbiological sciencesbiochemistrybiomoleculesproteins
- natural sciencesphysical sciencesopticsspectroscopyabsorption spectroscopy
- natural sciencescomputer and information sciencesartificial intelligencemachine learningdeep learning
- natural sciencesbiological sciencesmolecular biologystructural biology
Programma(i)
Argomento(i)
Meccanismo di finanziamento
MSCA-IF-EF-ST - Standard EFCoordinatore
8092 Zuerich
Svizzera