A0299
Title: Singular value decomposition to deconvolute mixed signals
Authors: Zhong Ren - University of Illinois Chicago (United States) [presenting]
Abstract: The experimental data often contain mixed signals from multiple causes of physical or chemical origins as we intentionally perturb a biological macromolecule in the hope of studying its structural and spectroscopic responses. The general approach is to gather a large amount of structural or spectroscopic data under a variety of experimental conditions or metadata. In the top-ranked isotropic Euclidean subspace established by singular value decomposition (SVD) of the large dataset of mixed signals, a multi-dimensional rotation that achieves a deconvolution of at least one pure signal is recently suggested. A pure signal refers to either a complicated structural response that is induced by a single factor or one isolated structural event that could be repeatedly triggered by different physical or chemical causes. A natural coordinate system may be further suggested by de-orthogonalization in the Euclidean subspace of SVD, which could achieve simultaneous deconvolution of several entangled signals. Examples from the research in structural biology are demonstrated.
