Speaker
Description
To understand how the conformational space of small peptide fragments determine the formation of protein three-dimensional structures is one of the important goals of modern biochemistry and structural biology. Achieving this goal will allow significant progress, for example, in the design of specific enzyme-based catalysts which would greatly simplify chemical synthesis in industry. Full control of the conformational behavior of protein fragments may also represent a way how to significantly deepen our understanding of protein folding and protein-ligand interactions.
To reliably elucidate the key determinants of conformational spaces of peptide fragments, we needed an extensive dataset. We used data from Peptide Conformational Samples dataset (PeptideCS), developed in our group, consisting of over 400 milions of dipetide off-equilibrium structures and 100 milions equilibrium structures, all computed at IT4I. The dataset contains structure, QM energy, QM energy in solvent (water), and gradients. To eliminate the redundancy in the equilibrium structures Because we found (equilibrium) structures of local minima to be redundant, we used machine learning clustering algorithm to select a set of unique representative minimas.
Enormous set of data represented by the PeptideCS database enabled us to investigate the impact of various physico-chemical factors determining the properties of the conformational space and of the minima (equilibrium structures) found. The factors include charge of the chain charge of the particular residue, its sterical requirements, and backbone-side chain interaction. These factors determine the shape of the conformational energy window, energy and histograms. In addition, we analysed and compared the energy surfaces of the dipeptides, defined by the 400 million off-equilibrium structures and investigated the secondary structure preferences, including energy barriers in the forbidden region of the Ramachandran diagram.
The extensive study should considerably improve the prediction of three-dimensional protein structure from the first principles (ab initio) which may complement recent successful efforts of the AI algorithms, such as AlphaFold2.
