 |
General: 663 |
P5: Assessing a Nonlinear Dimensionality Reduction-based Approach to Biological Network Reconstruction
Author: Vinodh N. Rajapakse , Advisor: Wojciech Czaja (Mathematics and Norbert Wiener Center)
Problem Statement Presentation
Project Proposal
Abstract
This project will build a basic, integrated data analysis pipeline for deriving gene association network models from gene expression data. Dimensionality reduction techniques will be applied to map the input data in ways that aim to capture intrinsic structure. After this step, standard network reconstruction and analysis techniques will be applied. Algorithm implementations will be individually validated using well-characterized data sets and established software. Following this foundational work, the impact of dimensionality reduction on the overall network reconstruction will be systematically assessed using additional validation and testing data sets. In particular, the initial aim will be to implement, validate, and incorporate a representative nonlinear dimensionality reduction technique – Laplacian Eigenmaps. Network reconstructions derived using data processed by this technique will be compared with ones derived using a leading method, with ones derived directly from the original data, as well as with ones derived from data processed using a standard linear dimensionality reduction approach – Principal Component Analysis. Pending successful completion of this work, some extensions of the basic analysis pipeline are possible. These notably include an approach to enhance handling of very large data sets, as well as consideration of an additional nonlinear dimensionality reduction technique – Diffusion Maps.
MidYear Progress Report and
Presentation
Final Presentation ,
Final Report
