Doubly Penalized LASSO for Reconstruction of Biological Networks

Reconstruction of biological and biochemical networks is a crucial step in extracting information from a large volume of biological data. There are several methods developed recently to reconstruct biological networks using dynamic data, each with specific benefits and some drawbacks. Here, we have developed a new method called Doubly Penalized Linear Absolute Shrinkage and Selection Operator (DPLASSO) for reconstruction of dynamic biological networks. In this approach, we have integrated two distinct methods viz., statistical significance testing of model coefficients and penalized/constrained optimization. Principal component analysis with statistical significance testing acts as a supervisory-level filter to extract the most informative components of the network from a dataset (Layer 1). In the lower level (Layer 2), LASSO with extra weights on the smaller parameters obtained in the first layer is employed to retain the main predictors and to set the small coefficients to zero. Two case studies are used to compare the relative performance of DPLASSO and LASSO in terms of several metrics, such as sensitivity, specificity, accuracy and fractional-error in the estimates of the coefficients. In the first case study, with a synthetic data set, our simulation results show substantial improvements over LASSO for the reconstruction of the network in terms of accuracy and specificity. The second case study relies on experimental datasets for cell division cycle of fission yeast. This case study illustrates that DPLASSO performs better than LASSO in terms of sensitivity, specificity and accuracy in reconstructing networks.