Reverse engineering methodology for bioinformatics based on genetic programming, differential expression analysis and other statistical methods

Abstract

This paper presents a robust automatic modelling of microRNA (miRNA) dynamics which has at core a Genetic Programming (GP) Reversing Ordinary Differential Equations Systems (RODES) methodology which was developed before and which is enhanced herein and consists of four steps: (1) smooth and fit the miRNA experimental data which is enhanced in this paper with other types of statistical analyses such as gene differential expression analyses, (2) decomposition of the transcription network or ODEs system, (3) automatically discovering the structure of networks or their ODEs system model by using GP and automatically estimating parameters of the ODEs systems models, (4) identification of the biochemical and pharmacological mechanisms. All four steps are paramount to the GP RODES methodology, which has been already applied in bioinformatics, while in this paper it is underlined a robust set of procedures for implementing the first step of the four step GP RODES framework together with an application of the GP RODES on a real miRNA dataset. Specifically it is highlighted a robust method to noise for fitting omics experimental input data, which consists of the Smoothing Spline Regression (SSR) algorithm based on the Generalized Cross Validation (GCV) criterion. The differential expression analysis of a dataset of 837 mirRNAs genes (GEO accession: GSE35074) is achieved by using various statistical methods. Furthermore, a GP algorithm (i.e. GP TIPS software) is used on the SSR and GCV fitted miRNA200a (microRNA200a), miRNA200b and miRNA424, which were selected from a larger group of identified differential expressed miRNA genes and with the scope of predicting accurately the miRNAs derivatives with regard to time. While the computational findings are in agreement with the ones from the initial study by Moes et al. (2012) [21], this paper develops an enhanced GP RODES methodology which can be further applied to bioinformatics. © 2014 Springer International Publishing Switzerland.