Efficient and Robust Algorithms for Statistical Inference in Gene Regulatory Networks

Abstract

Inferring gene regulatory networks (GRNs) is of profound importance in the field of computational biology and bioinformatics. Understanding the gene-gene and gene- transcription factor (TF) interactions has the potential of providing an insight into the complex biological processes taking place in cells. High-throughput genomic and proteomic technologies have enabled the collection of large amounts of data in order to quantify the gene expressions and mapping DNA-protein interactions.

This dissertation investigates the problem of network component analysis (NCA) which estimates the transcription factor activities (TFAs) and gene-TF interactions by making use of gene expression and Chip-chip data. Closed-form solutions are provided for estimation of TF-gene connectivity matrix which yields advantage over the existing state-of-the-art methods in terms of lower computational complexity and higher consistency. We present an iterative reweighted ℓ2 norm based algorithm to infer the network connectivity when the prior knowledge about the connections is incomplete.

We present an NCA algorithm which has the ability to counteract the presence of outliers in the gene expression data and is therefore more robust. Closed-form solutions are derived for the estimation of TFAs and TF-gene interactions and the resulting algorithm is comparable to the fastest algorithms proposed so far with the additional advantages of robustness to outliers and higher reliability in the TFA estimation.

Finally, we look at the inference of gene regulatory networks which which essentially resumes to the estimation of only the gene-gene interactions. Gene networks are known to be sparse and therefore an inference algorithm is proposed which imposes a sparsity constraint while estimating the connectivity matrix.The online estimation lowers the computational complexity and provides superior performance in terms of accuracy and scalability.

This dissertation presents gene regulatory network inference algorithms which provide computationally efficient solutions in some very crucial scenarios and give advantage over the existing algorithms and therefore provide means to give better understanding of underlying cellular network. Hence, it serves as a building block in the accurate estimation of gene regulatory networks which will pave the way for finding cures to genetic diseases.