Navržení booleovské sítě na základě dat genové exprese u nemodelových organismů

Abstract

Boolean network inference has emerged as an important method for studying gene regulation processes in biological systems. This thesis develops a comprehensive pipeline for inferring genome-wide Boolean networks from RNA-Seq data in the non-model organism Clostridium beijerinckii NRRL B-598. The main aim of this research is to construct a complete regulatory network without the need for pre-existing biological data. The study aims to identify regulatory connections involved in solvent production through an assessment of gene expression patterns across different experimental conditions. This network is built in two stages: R-based preprocessing for fractional counting of multi-mapped reads and Python-based Boolean rule inference with decision tree classifiers. The Bioconductor tools were used for alignment processing, while scikit-learn and bespoke algorithms were employed to build the Boolean rules. The network analysis results demonstrated outstanding performance with successful inference for all 5530 genes, resulting in 17935 regulatory edges, 99.98% of which satisfied high-quality metrics. Notably, the network has biologically plausible features such as sparse connectivity (density = 0.00059) and a high proportion of activation links (97.2%) indicating coordinated regulatory mechanisms. Nonetheless, acknowledging some limitations of this study is essential. The binary discretization of continuous expression data, along with the analysis of just one experimental condition, can risk oversimplifying complex regulatory mechanisms. Moreover, the absence of strict experimental validation that defines research on non-model organisms obstructs biological confirmation of proposed correlations. This limitation highlights the necessity of future experimental validation to verify the computational predictions. This research lays the groundwork for a genome-wide Boolean regulatory network for C. beijerinckii and offers a scalable framework that can be used with other non-model organisms.Boolean network inference has emerged as an important method for studying gene regulation processes in biological systems. This thesis develops a comprehensive pipeline for inferring genome-wide Boolean networks from RNA-Seq data in the non-model organism Clostridium beijerinckii NRRL B-598. The main aim of this research is to construct a complete regulatory network without the need for pre-existing biological data. The study aims to identify regulatory connections involved in solvent production through an assessment of gene expression patterns across different experimental conditions. This network is built in two stages: R-based preprocessing for fractional counting of multi-mapped reads and Python-based Boolean rule inference with decision tree classifiers. The Bioconductor tools were used for alignment processing, while scikit-learn and bespoke algorithms were employed to build the Boolean rules. The network analysis results demonstrated outstanding performance with successful inference for all 5530 genes, resulting in 17935 regulatory edges, 99.98% of which satisfied high-quality metrics. Notably, the network has biologically plausible features such as sparse connectivity (density = 0.00059) and a high proportion of activation links (97.2%) indicating coordinated regulatory mechanisms. Nonetheless, acknowledging some limitations of this study is essential. The binary discretization of continuous expression data, along with the analysis of just one experimental condition, can risk oversimplifying complex regulatory mechanisms. Moreover, the absence of strict experimental validation that defines research on non-model organisms obstructs biological confirmation of proposed correlations. This limitation highlights the necessity of future experimental validation to verify the computational predictions. This research lays the groundwork for a genome-wide Boolean regulatory network for C. beijerinckii and offers a scalable framework that can be used with other non-model organisms.