High-grade serous ovarian cancer (HGSOC) is initially responsive to platinum-based chemotherapy, but most patients eventually relapse with treatment-resistant disease. Single-cell RNA-sequencing (scRNA-seq) provides a powerful tool to study intratumoral heterogeneity and uncover resistant cancer cell populations, yet its analysis requires tailored computational methods. This thesis first explores CA-125-based chemosensitivity scores as alternatives to the platinum-free interval (PFI), a widely used measure of resistance that can be confounded by factors such as follow-up variability. The analysis shows that CA-125-based scores correlate with PFI in neoadjuvant-treated patients but perform less consistently across other treatment settings, underscoring the need for more robust metrics. To identify resistant cell subpopulations and guide treatment selection, the thesis then introduces a bioinformatics pipeline that combines scRNA-seq, clinical data, and large-scale drug screens. Using non-negative matrix factorization (NMF), the pipeline decomposes gene expression into signatures, which are assessed for their association with treatment resistance. Applied to a scRNA-seq dataset of 57 tumor samples from 42 HGSOC patients collected before, during, and after chemotherapy, the analysis reveals resistance-linked signatures and predicts candidate drugs, several of which were validated in patient-derived organoids and xenografts. Finally, a supervised extension of NMF is proposed to better detect phenotype-associated expression programs by integrating patient-level resistance annotations into the factorization process. This approach improves the identification of relevant signatures in simulated data, as well as in a clinical scRNA-seq dataset of HGSOC.