The design of effective combination therapies for cancer depends on identifying the druggable signaling pathways that transformed cells use to make phenotypic decisions. Finding these pathways is challenging because signaling information is high-dimensional and because cancer cells exhibit substantial cell-to-cell variability and phenotypic plasticity in the heterogeneous tumor microenvironment. Overcoming these barriers requires the careful integration of appropriate computational models with experimental data that describe the complexity of tumor biology at multiple scales. This talk will focus on our integration of data science with mechanistic modeling approaches to understand how pancreas cancer cells make decisions leading to chemoresistance in response to diverse cues in the tumor microenvironment. Models are trained on biochemical measurements and high-content imaging that capture the dynamics and heterogeneity of the multivariate signaling processes leading to cell phenotype determination. Model predictions are tested in mouse models and cell culture experiments and are further validated through the analysis of human patient data. Our results nominate specific candidate drug combinations as potentially more effective pancreas cancer treatments and demonstrate how cancer systems biology approaches can be successfully deployed for the effective design of preclinical and clinical studies.