Regularization as a Toolkit for Parsimonious Modeling in Bioinformatics

With the tremendous progress in bioinformatics we are facing new challenges where complex data set need to be analyzed to glean meaningful scientific knowledge. Though statistical tools were slow to emerge in the beginning, however it was discovered early that there exist some seemingly related issues in many of the different bioinformatics data sets. They are generally overly complex giving rise to complicated models which are often sub-optimal and have poor generalization capability outside the realm of respective study. This violates scientific principles of causality. The balance between the model complexity and its generalizing capacity is often sought after in many of the early successful modeling efforts in bioinformatics. Over the time this has been termed as “Regularization”. This review article investigates regularization as a solution for parsimonious model selection in complex biological data and can be seen as an attempt to gather bits and pieces of many such earlier efforts under a single umbrella.