Statistical Analysis of high dimensional molecular data

CREATED: 200906250534 Speaker: William R. Atchley Web: http://www4.ncsu.edu/~lgmcferr/ICMSBWorkshop/Home.html ** Multivariate statistical procedures

Cluster analysis
PCA, ICA
Factor analysis
Latent variable analysis
main objective is to reduce dimensionality in a meaningful way ** Cluster Analysis
ubiquitous in microarray studies
not much supporting statistical theory ** PCA
dimension reduction by optimizing variance explanation
ordination
with noisy data, principle components may reflect stochastic effects and environmental noise ** Latent variable model
relates a set of observed or manifest variables to a set of unobserved or latent variables
latent variables often reflect phenomenon of interest
separates variability that is shared (commonality) or unique
latent variables by factor analysis
exploratory factor analysis: detect natural groupings of variables (factors) based on shared variability
e.g. Latent structure of AA physiochemical variation (Atchley2005, PNAS) ** Utility of Factor Scores
each amino acid can be represented by factor scores for each latent variable, summarizes almost 500 physiochemical attributes
new way of aligning protein sequences base on factor scores
factor scores as data in discriminant analysis (Atchley2006)