CF-GeNe: Fuzzy Framework for Robust Gene Regulatory Network Inference

Abstract

Most Gene Regulatory Network (GRN) studies ignore the impact of the noisy nature of gene expression data despite its significant influence upon inferred results. This paper presents an innovative Collateral-Fuzzy Gene Regulatory Network Reconstruction (CF-GeNe) framework for Gene Regulatory Network (GRN) inference. The approach uses the Collateral Missing Value Estimation (CMVE) algorithm as its core to estimate missing values in microarray gene expression data. CF-GeNe also mimics the inherent fuzzy nature of gene co-regulation by applying fuzzy clustering principles using the well-established fuzzy cmeans algorithm, with the model adapting to the data distribution by automatically determining key parameters, like the number of clusters. Empirical results confirm that the CMVE-based CF-GeNe paradigm infers the majority of co-regulated links even in the presence of large numbers of missing values, compared to other data imputation methods including: Least Square Impute (LSImpute), K-Nearest Neighbour Impute (KNN), Bayesian Principal Component Analysis Impute (BPCA) and ZeroImpute. The statistical significance of this improved performance has been underscored by gene selection and also by applying the Wilcoxon Ranksum Significance Test, with results corroborating the ability of CF-GeNe to successfully infer GRN interactions in noisy gene expression data.