Year: 2009
Type: dissertation
Abstract: Traditional protein tertiary structure prediction methods have focused on predicting structures using a combination of alignment algorithms, scoring functions, potential-energy function, and large-scale conformation search. By contrast, we propose, instead, to identify a simplified version of the dynamics involved in the folding process and use these dynamics to predict the folded state of proteins with unknown structure. A machine learning system is developed to induce the folding dynamics of protein structures from linear amino acid sequences with known structures (training set). These dynamics are then applied to other unseen linear amino acid sequences whose structures are unknown (test set). Thus, the problem of identifying the fold dynamics is treated as a function approximation problem, where the function to be learned is the update equation of a dynamical system. This novel approach represents a middle ground between homology modeling and 'ab-initio' methods. Rather than explicitly searching for a matching structure to predict an unknown structure's configuration, it uses interpolation and extrapolation between the structures of the training set to infer the appropriate dynamics without an explicit physics model. We have induced three-dimensional models of the backbone (Amine-Nitrogen ('N'), Alpha-Carbon ('C' [alpha]), and Carboxyl-Carbon ('C')) for protein fragments of 16 residues long and their accuracies are comparable to those of threading methods (i.e. about 3-5A? deviation from the resolutions of the corresponding X-ray structure). However, unlike threading methods, the proposed method is able to induce these models without requiring knowledge of the remaining protein structure. Specifically, current methods require highly accurate information about the remaining protein structure in order to perform prediction of small sub-structures of interest.
Author: Lac, Hao Chi
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W6991857985