Computational approaches for the identification of candidate chemotheraphy-related lncRNAs in HGSOvCa

High grade serous ovarian cancer (HGSOvCa) is a malignant tumor subtype that originates from the female reproductive system. The standard therapies prescribed to HGSOvCa patients include several chemotherapy cycles based on platinum-taxol drugs and a debulking surgery for removing the cancer tissues. A fundamental characteristic of this disease, that drastically decreases the 5-years survival rates, is the acquisition of chemotherapy resistance by the tumoral cells after the first-line treatment. Both the cancer aggressiveness and the development of the platinum resistance increase the necessity of a more effective and targeted therapy.During the last 10 years, a branch of the cancer research has focused its attention on the genomic components called "long non-coding RNAs". These elements, originating from RNA molecules, do not encode for proteins and are composed by a number of nucleotides that ranges from 200 to 100000. Even if they do not have encoding properties, it was shown that those transcripts are actively involved in many cell functions and they are dysregulated during the genesis and the development of different tumors. The main goal of this master thesis is to develop a pipeline for the automatic identification of long non-coding RNAs that can be possibly involved in the platinum-resistance process (generally called drivers). By knowing the drivers and the molecular processes that lead to chemotherapy resistance it would be possible to identify efficient pharmacological targets and design a more effective therapy. This thesis was conducted in collaboration with the System Biology Lab for Drug Resistance of the Helsinki University in Finland. The data employed in the analysis were measured amount of different long non-coding RNA molecules (called expression levels) obtained through the total RNA-sequencing of different HGSOvCa patients’ samples, collected when the disease is diagnosed. In order to achieve the proposed goal, the analysis was focused on the identification of genes that showed different behaviours in the chemo-resistant patients with respect to the chemo-sensitive ones. For this reason, samples were initially divided in two groups, according to the available clinical data. The analysis was conducted by realizing a pipeline that integrates two different strategies: an unsupervised hierarchical clustering approach supported by statistical processing and a supervised procedure composed by two feature selection methods based on the Random Forest and the Boruta algorithms. The choice of combining different methods for the same analysis was raised by the necessity of having more confident results. The outcomes of the two employed methods, in fact, were finally compared and only the long non-coding RNAs identified by all the techniques were taken into account. From the analysis were retrieved 6 long non-coding RNAs that can potentially be related to the chemotherapy resistance and that are currently under validation. In this work of thesis were also highlighted the strength and the weaknesses of the adopted approaches as well as the limitations of the study encountered during the analysis.