New Algorithms for Predicting Conformational Polymorphism and Inferring Direct Couplings for Side Chains of Proteins
Year: 2015
Type: dissertation
Abstract: Protein crystals populate diverse conformational ensembles. Despite much evidence that \nthere is widespread conformational polymorphism in protein side chains, most of the xray \ncrystallography data are modelled by single conformations in the Protein Data Bank. \nThe ability to extract or to predict these conformational polymorphisms is of crucial importance, \nas it facilitates deeper understanding of protein dynamics and functionality. \nThis dissertation describes a computational strategy capable of predicting side-chain polymorphisms. \nThe applied approach extends a particular class of algorithms for side-chain \nprediction by modelling the side-chain dihedral angles more appropriately as continuous \nrather than discrete variables. Employing a new inferential technique known as particle \nbelief propagation (PBP), we predict residue-speci c distributions that encode information \nabout side-chain polymorphisms. The predicted polymorphisms are in relatively close \nagreement with results from a state-of-the-art approach based on x-ray crystallography \ndata. This approach characterizes the conformational polymorphisms of side chains using \nelectron density information, and has successfully discovered previously unmodelled \nconformations. \nFurthermore, it is known that coupled \nuctuations and concerted motions of residues \ncan reveal pathways of communication used for information propagation in a molecule \nand hence, can help in understanding the \\allostery" phenomenon in proteins. In order \nto characterize the coupled motions, most existing methods infer structural dependencies \namong a protein's residues. However, recent studies have highlighted the role of coupled \nside-chain \nuctuations alone in the allosteric behaviour of proteins, in contrast to a \ncommon belief that the backbone motions play the main role in allostery. These studies \nand the aforementioned recent discoveries about prevalent alternate side-chain conformations \n(conformational polymorphism) accentuate the need to devise new computational \napproaches that acknowledge side chains' roles. As well, these approaches must consider \nthe polymorphic nature of the side chains, and incorporate e ects of this phenomenon \n(polymorphism) in the study of information transmission and functional interactions of \nresidues in a molecule. Such frameworks can provide a more accurate understanding of the \nallosteric behaviour. \nHence, as a topic related to the conformational polymorphism, this dissertation addresses \nthe problem of inferring directly coupled side chains, as well. First, we present a \nnovel approach to generate an ensemble of conformations and an e cient computational \nmethod to extract direct couplings of side chains in allosteric proteins. These direct couplings \nare used to provide sparse network representations of the coupled side chains. The \nframework is based on a fairly new statistical method, named graphical lasso (GLASSO), \niii \ndevised for sparse graph estimation. In the proposed GLASSO-based framework, the sidechain \nconformational polymorphism is taken into account. It is shown that by studying \nthe intrinsic dynamics of an inactive structure alone, we are able to construct a network of \nfunctionally crucial residues. Second, we show that the proposed method is capable of providing \na magni ed view of the coupled and conformationally polymorphic side chains. This \nmodel reveals couplings between the alternate conformations of a coupled residue pair. To \nthe best of our knowledge, this is the rst computational method for extracting networks \nof side chains' alternate conformations. Such networks help in providing a detailed image \nof side-chain dynamics in functionally important and conformationally polymorphic sites, \nsuch as binding and/or allosteric sites. This information may assist in new drug-design \nalternatives. \nSide-chain conformations are commonly represented by multivariate angular variables. \nHowever, the GLASSO and other existing methods that can be applied to the aforementioned \ninference task are not capable of handling multivariate angular data. This dissertation \nfurther proposes a novel method to infer direct couplings from this type of data, and \nshows that this method is useful for identifying functional regions and their interactions in \nallosteric proteins. The proposed framework is a novel extension of canonical correlation \nanalysis (CCA), which we call \\kernelized partial CCA" (or simply KPCCA). Using the \nconformational information and \nuctuations of the inactive structure alone for allosteric \nproteins in the Ras and other Ras-like families, the KPCCA method identi ed allosterically \nimportant residues not only as strongly coupled ones but also in densely connected \nregions of the interaction graph formed by the inferred couplings. The results were in good \nagreement with other empirical ndings and outperformed those obtained by the GLASSO-based framework. By studying distinct members of the Ras, Rho, and Rab sub-families, \nwe show further that KPCCA is capable of inferring common allosteric characteristics in \nthe small G protein super-family.
Source: UWSpace (University of Waterloo)
Author: Laleh Soltan Ghoraie
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W1885077158