Conditional Random Fields with Least Absolute Shrinkage and Selection Operator to Classifying the Barley Genes Based on Expression Level Affected by the Fungal Infection

Abstract

The classical methods for the classification problem include hypothesis test with the Benjamini–Hochberg method, hidden Markov chain model, and support vector machine. One major application of the classification problem is gene expression analysis, for example, detecting the host genes having interaction with pathogen. The classical methods can be applied and have a good performance when the number of genes having interaction with the pathogen is not sparse with respect to the candidate genes. However, conditional random field (CRF), with an appropriate design, can be applied and have good performance even when it is sparse. In this work, we proposed a modified CRF with a baseline to reduce the number of parameters in CRF. Moreover, we show an application of CRF with the least absolute shrinkage and selection operator (LASSO) to classifying barley genes of its reaction to the pathogen.