Partial least squares method for the analysis of gene expression data

Partial Least Squares Regression (PLSR) is an unsupervised machine learning technique to modeling associations between variables through orthogonal latent variables. Using these latent variables, PLSR can make an inference from huge and computationally complex datasets that have missing values, noise and a numerous number of variables relativity more than the number of observations. The classical and standard algorithm of the PLSR is the Nonlinear Iterative Partial Least Squares Regression (NIPALS). The NIPALS is proposed for regression, classification and dimension reduction. The NIPALS and other PLSR algorithms have been used frequently for various bioinformatic studies. In high-throughput gene expression data research, one of the important goals is to investigate gene-gene or their products interactions. To measure the level of association between these genes or their products, a highly recommended method can be used which is calculated by the variable weights and loadings based on PLSR, called Connectivity Scores. In this thesis, PLSR was used for computing connectivity scores to construct gene networks for three brain region of a developing mouse brain in the embryonic period. Statistical analysis is performed using R statistical language and Cytoscape software is used to visualize gene networks.