A data-informed approach for engineering in-vitro experiment design to decipher key features of invasive breast cancer cell phenotypes

The intrinsic complexity of biological processes often hides the role of dynamic microenvironmental cues in the development of pathological states. The use of micro-physiological systems (MPS) offers new technological platforms designed to model the dynamics of tissue-specific microenvironments in-vitro and to holistically understand healthy and pathological states. In our previous works, we reported on engineering breast critical tumor microenvironment features, including matrix stiffness, pH, and fluid flow, and use the MPSs to study breast cancer cells phenotypes. By studying different microenvironments mimicking normal and tumor breast tissues, we obtained high-dimensional data using two distinctive human breast cell lines (i.e., MDA-MB231, MCF-7) investigating biomarkers commonly used in cancer in-vitro models as cell proliferation, epithelial-to-mesenchymal transition (EMT), and breast cancer stem cell markers (B-CSC). We herein report on a new approach used to explore the complexity of MPSs and the high dimensional datasets: we introduce an innovative machine learning (ML) based platform employing unsupervised k-means clustering and feature extraction to identify key markers that differentiated invasive from non-invasive breast cell phenotypes. This novel data-driven approach streamlines experimental design and emphasizes the translational potential of integrating MPS-derived insights with ML to refine prognostic tools and personalize therapeutic strategies