Metagenomics allows for a research of microorganisms which cannot be cultivated and explores them as entire microbial communities. Among many research directions that are opened through metagenomics, these studies also contribute to the investigation of human microbiome with the aim of discovering causes for different diseases, as well as inventing the methods for their diagnosis and treatment. Prokaryotes do not use all synonymous codons with the same frequency and those genes which are coded with preferred codons are optimised for translation. This effect is termed translational optimisation. By comparing individual genes to highly expressed gene set, such as ribosomal protein genes, we can predict their expressivity – their potential to be expressed. It has been demonstrated previously that this is an effect that can be observed not only at the level of a single species, but also at the level of entire microbial communities. Translational optimisation effect was used in this thesis as a basis for building a classification model based on a random forest algorithm. By training the model on the observed differences in codon usage bias between microbial communities from healthy and liver diseased individuals, we can predict with high accuracy the disease status in unknown samples. Also, an exploratory analysis of an entire dataset was performed to examine genes that are most important for the classification of samples. Pathway analysis was conducted to examine pathways in which the samples vary and to identify potential biomarkers for diagnosis of liver disease. Additional physiological information about the patients was observed in search for a correlation with the condition of samples.