Thesis topic proposal
Tibor Páli
Membrane protein structure prediction by combining machine learning and molecular mechanics methods


Institute: University of Szeged
Doctoral School of Physics

Thesis supervisor: Tibor Páli
Location of studies (in Hungarian): ELKH BRC
Abbreviation of location of studies: SZBK

Description of the research topic:

In the era of proteomics, complete genome databases, it is much easier to assign protein sequences to biological functions than just a decade ago. The vast majority of proteins assume a native structure in their native environment for proper functioning. The native structure changes during the biological activity but the mean conformation is nevertheless characteristic of the given protein. Structural biology aims at determining the structure of bio-molecules at various spatial and temporal resolutions. The experimental efforts rely mostly on classical structural biology techniques, namely X-ray crystallography and NMR structure determination. However, classical structure biology of membrane proteins is alarmingly behind that of soluble proteins, because it requires that membrane proteins be taken out from their native environment, the lipid bilayer, and then be either crystallised or solubilised in a folded form. In addition, the biological, functional relevance of these structures are always subject to debate. It is therefore not surprising that out of the more than 100 thousand known protein structures there are only a few hundreds of trans-membrane membrane proteins, whereas roughly 30% of the genes in the human genome code such proteins. There is a huge interest in membrane protein structures and in alternative approaches. In the past, using molecular mechanics and structural data from spectroscopy, we have "manually" modeled the membrane location of cytochrome c, based on our own proximity measurements in reconstituted bilayer. The backbone structures of the major coat protein of the M13 bacteriophage were identified that are most compatible with structural data (side chain mobility and membrane location) measured in membranes. We have also verified structures in detergent vesicles (determined by NMR) under the constraints of functionally relevant structural data. Combining bioinformatics, sequence-based predictions, loop databases and molecular mechanics, we predicted the basic fold of the conserved four-helix trans-membrane bundle of the cytochrome b561 membrane protein family. We have also established the theoretical framework for obtaining the orientation of trans-membrane alpha-helix and beta-barrel bundles in a phospholipid bilayer from their polarised attenuated total reflection (ATR) FTIR spectra, which we then applied for gramicidin A and polypeptides related to V-ATPase. The main objective of this project is to improve existing protein structure prediction techniques, involving modern prediction approaches and experimental constraints, with focus on membrane proteins. We will achieve this objective through the following steps:
• Neuronal network methods will be implemented, refined and trained for predicting structural features from a sequence of a protein of unknown structure. Initially we will restrict the study to homologous membrane proteins.
• The predicted structural features will be converted to mechanical constraints and applied in a conformational search based on molecular mechanics potentials at residue-level resolution.
• The structures of lowest virtual free energy will be selected and further minimised in all-atom molecular mechanics simulations. In these simulations experimantal constraints will be also added if either available from the literature or measured in the group for the given protein.

Required language skills: English
Number of students who can be accepted: 2

Deadline for application: 2023-06-30

All rights reserved © 2007, Hungarian Doctoral Council. Doctoral Council registration number at commissioner for data protection: 02003/0001. Program version: 2.2358 ( 2017. X. 31. )