Computational methods to analyze biological networks from transcriptomics data

The regulation of biological processes within each single cell governs all the main cellular mechanisms aimed at the development, differentiation, and proper maintenance of the cell itself. It determines, at the tissue level and more generally at the organism level, the actual role assumed by the cell. It is of vital interest to the scientific and biological community to establish precisely how such regulatory processes are concretized in various cell types, whether they are considered healthy or diseased. Understanding these mechanisms necessarily involves identifying and sequencing the gene products that each cell synthesizes and possesses at a given moment. The technique at the core of this cellular study is called single-cell analysis, which is capable of generating multi-omics data that represent all the main regulatory mechanisms occurring within the isolated and analyzed cells. This technology can be considered as the basis for a more accurate study of the roles' heterogeneity exploited by cells within their respective tissues. In this regard, an effective representation of such processes is provided by gene regulatory networks, a tool of particular interest to the biological and scientific community. However, the inference of such networks remains an unresolved issue, with methods that are not simultaneously effective and efficient. This situation is further complicated by the need to study data from single-cell analyses and multi-omics data, which have a significantly larger dimension than the sequencing data obtained from previous generation technologies. This work aims to analyze two bioinformatics methods for the inference of gene regulatory networks, highlighting their biological potentials and computational limits. Of particular interest, in evaluating these methods for real applicability in experiments with single-cell and multi-omics data, is the scalability of the processes used by such software. This property is also evaluated based on the possibilities for parallelization that these software packages present already in their implementation. This analysis aims to evaluate potential computational improvements, also utilizing the capabilities of parallel computing on GPU architectures. The main tools used to implement these modifications are software libraries considered state-of-the-art in parallel computing, coupled with GPU structures capable of fully exploiting their potential. The improved versions of the two bioinformatics methods show significantly reduced execution times compared to those obtained from the original version, with comparable use of memory and resources.