Biancaniello, Carmen (2024) NapShift: a neural network-based method to predict NMR chemical shifts and improve biomolecular simulations. [Tesi di dottorato]
Preview |
Text
Biancaniello_Carmen_36.pdf Download (30MB) | Preview |
| Item Type: | Tesi di dottorato |
|---|---|
| Resource language: | English |
| Title: | NapShift: a neural network-based method to predict NMR chemical shifts and improve biomolecular simulations. |
| Creators: | Creators Email Biancaniello, Carmen carmen.biancaniello@unina.it |
| Date: | 11 March 2024 |
| Number of Pages: | 146 |
| Institution: | Università degli Studi di Napoli Federico II |
| Department: | Ingegneria Elettrica e delle Tecnologie dell'Informazione |
| Dottorato: | Computational and quantitative biology |
| Ciclo di dottorato: | 36 |
| Coordinatore del Corso di dottorato: | nome email Ceccarelli, Michele michele.ceccarelli@unina.it |
| Tutor: | nome email De Simone, Alfonso UNSPECIFIED |
| Date: | 11 March 2024 |
| Number of Pages: | 146 |
| Keywords: | Proteins, artificial neural networks, chemical shifts, hybrid potential, molecular dynamics, basin-hopping global optimization. |
| Settori scientifico-disciplinari del MIUR: | Area 05 - Scienze biologiche > BIO/11 - Biologia molecolare |
| Additional information: | sede di lavoro: Dipartimento di Farmacia |
| Date Deposited: | 19 Jun 2024 12:21 |
| Last Modified: | 09 Mar 2026 13:33 |
| URI: | http://www.fedoa.unina.it/id/eprint/15417 |
Collection description
Computational methods used to investigate biomolecule structure and dynamics are valuable approaches to overcome the intrinsic limitations of experimental techniques. The implementation of hybrid restraint potential that incorporates experimental data to refine molecular mechanics force fields on which these methods rely, has been proved to enhance various computational methodologies, such as molecular dynamics, structure prediction, and energy landscape sampling. This study embarks on the development of a novel hybrid restraint potential founded on NapShift, an artificial neural network that predicts protein nuclear magnetic resonance (NMR) chemical shifts from protein sequence and structure. NapShift achieves accurate predictions of chemical shifts, and its key feature lies in its full differentiability with respect to atomic coordinates, enabling its application in structural refinement. Utilizing NapShift throughout simulations allows for the computation of an energy penalty based on the difference between experimental chemical shifts and those predicted from the conformation obtained at each simulation step. This additional term provides differentiated hybrid restraint forces to complement empirical force fields. The NapShift hybrid restraint potential was subjected to rigorous evaluation through basin-hopping global optimization and molecular dynamics simulations performed on various protein systems, comparing the original and hybrid force fields. The integration of NapShift demonstrated to remarkably enhance the performance of both simulation approaches, resulting in better structure prediction with basin-hopping and increased local stability in molecular dynamics simulations. Additionally, the application of this methodology on challenging proteins revealed promising results in exploring protein structure-dynamics complex behavior using NMR chemical shifts. Overall, the outcomes of this study suggest that neural network hybrid potentials based on NMR observables can improve a wide range of molecular simulation techniques.
Downloads
Downloads per month over past year
Actions (login required)
 |
View Item |
