A Non-Negative Tensor Factorization Approach to Deconvolute Microenvironment in Breast Cancer

MotivationAn in-depth understanding of epithelial breast cell responses to the growth-promoting ligands is required to elucidate how the microenvironment (ME) signals affect cell-intrinsic regulatory networks and the cellular phenotypes they control, such as cell growth, progression, and differentiation. This is particularly important in understanding the mechanisms of breast cancer initiation and progression. However, the current mechanisms by which the ME signals influence these cellular phenotypes are not well established. ResultsWe developed a high-order correlation method using proteomics data to reveal the regulatory dynamics among proteins, histones, and six growth-promoting ligands in the MCF10 cell line. In the proposed method, the protein-ligand and histone-ligand correlations at multiple time points are first encoded in two three-way tensors. Then, a non-negative tensor factorization model is used to capture and quantify the protein-ligand and histone-ligand correlations spanning all time points, followed by a partial least squares regression process to model the correlations between histones and proteins. Our method revealed the onset of specific protein-histone signatures in response to growth ligands contributing to distinct cellular phenotypes that are indicative of breast cancer initiation and progression. Contactsmollah@wustl.edu