Reconstruction of Large-Scale Gene Regulatory Networks Using Regression-based Models

Gene regulatory networks (GRN) reconstruction is the process of identifying gene regulatory interactions from experimental data through computational analysis. GRN reconstruction-related works have boosted many major discoveries in finding drug targets for the treatment of human diseases, including cancer. However, reconstructing GRNs from gene expression data is a challenging problem due to high-dimensionality and very limited number of observations data, severe multicollinearity and the tendency of generating cascade errors. These problems lead to the reduced performance of GRN inference methods, hence resulting in the method being unreliable for scientific usage. We propose a method called P-CALS (Principal Component Analysis and Partial Least Squares) that is derived from the combination of PCA (Principal Component Analysis) with PLS (Partial Least Squares). The performance of P-CALS is assessed to the genome-scale GRN of E. coli, S. cerevisiae and an in-silico datasets. We discovered that P-CALS achieved satisfactory results as all of the sub-networks from diverse datasets achieved AUROC values above 0.5 and gene relationships were discovered at the most complex network tested in the experiments.