Multi-dimensional Components Identification and Estimation

Multiple-component analysis in high-dimensional regression models is a new research statistical problem that arose with recent technological advancements in biomolecular studies and proteomics. My primary goal in this research project is to develop rigorous statistical methods for multiple-component identification, estimation and inference through a unified statistical framework - multivariate additive regression. Integral to the methodological research is its application to biomolecular studies using Nuclear Magnetic Resonance (NMR).

A series of challenges arise in applications falling within the high-dimensional additive regression framework: inhomogeneous signal-to-noise ratio and possible artifacts confounded with the signal components; different trade offs between signal resolution and sensitivity across different data dimensionality; large number of components; partial or complete mixing of the components; and large data dimensionality. My research methods in this methodological research field address these difficulties with a solid statistical foundation and in a way which incorporates knowledge of the underlying study.

This web page has been initiated with the National Science Foundation support (DMS-1105191).

1. Serban, N. (2007), "MICE: Multiple-peak Identification, Estimation, and Characterization", Biometrics, 63,531-539.[.pdf ]

2. Serban, N. (2010), "Noise Reduction for Enhanced Component Identification in Multi-Dimensional Biomolecular NMR Studies", Computational Statistics and Data Analysis, 54 (4), pp 1051-1065.[.pdf ]

3. Serban, N., Li, P. (2014) "A statistical test for Mixture Detection in A Multi-dimensional Regression Model", Canadian Journal of Statistics, 42(1), 36-60.[.pdf ]

4. Hilton, R., Serban, N. (2014) "Theoretical Limits of Component Identification in a Separable Nonlinear Least Squares Problem", Journal of Nonparametric Statistics, 26(4), 769-791.
[Paper Document] [Software]

Back to Nicoleta Serbans Home page