Exploiting mechanistic models toward personalised strategies in Breast Cancer

Motivation. Breast cancer, the most prevalent malignancy among women globally, is an exceed- ingly clinically and genetically heterogeneous tumor type. It is impacted by interactions between cancer cells and cancer-associated fibroblasts (CAFs) within the tumor microenvironment, prompt- ing cancer cells to acquire a more aggressive mesenchymal phenotype. However, the mechanisms underlying this interaction remain to be elucidated. To fill thi gas we develepoped PatientProfiler, a computational pipeline able to concretely address how the genetic and molecular background of in- dividual patients contributes to the establishment of a malignant phenotype. As a proof of concept, we made use of publicly available multi-omic data derived from 122 treatment-na¨ıve breast cancer biopsies. Methods. To set up this approach we first developed PatientProfiler, a computational strategy that was designed to analyse patient derived data. PatientProfiler is composed of several functions that allow for multi-omic data analysis and standardization; generation of stratification groups; gen- eration of patient-specific models of signal transduction; extraction of prognostic biomarkers. To benchmark the tool, we retrieved genomic, transcriptomic, (phospho)proteomic and clinical data derived from 122 treatment-na¨ıve breast cancer biopsies, availble at the CPTAC portal. Next, we generated different stratification groups based on histopathologic classification of biopsies and func- tional activation of CAFs as derived from enrichment analysis on proteomic data. In parallel, other multi-omic data were used in input to the SignalingProfilier algorithm, which integrates experimen- tal data with prior knowledge from the literature to estimate protein activities from experimental data and to generate personalized mechanistic networks. These networks and patients stratifications have been used to identify, with a statistical approaches, specific network modules deregulated in patients belonging to specific subgroups. Results. Thanks to the development of PatientProfiler we have identified patient-specific mech- anistic models (one patient, one network) that recapitulate the deregulated signaling pathways in breast cancer, offering valuable insights into the underlying mechanisms of tumorigenesis and disease progression. Furthermore, our findings have facilitated the identification of potential biomarker pro- teins or modules, tailored to different stratification groups. In summary, through the integration of multi-omics data and computational techniques, we were able to shed light to the intricate interplay between cancer cells and CAFs, elucidating their role in driving cancer progression towards a more aggressive mesenchymal phenotype.