Advancements in omics technologies have promoted the development of precision oncology. Lineage plasticity, a hallmark of cancer, incorporates molecular and histological aspects. Histological differentiation of adenocarcinoma, neuroendocrine, and squamous characteristics occurs in different anatomic locations. Lung cancer, which is highly heterogeneous, encompasses these differentiations, and therefore serves as a model for exploration. Data-driven understanding is critical in cancer differentiation research, with the two major differentiation pathways, squamous and neuroendocrine, supported by omics data. Here, genetic and non-genetic profiles are reviewed based on patient datasets, and shareable molecular features are described. This paper mainly discusses machine learning approaches to feature selection, where network modeling is effective for designing programmable differentiation. All methods are presented within the context of cancer lineage plasticity along with examples and hypotheses. It emphasizes that selected patient datasets combined with methods will ultimately lead to actionable cancer lineage. Chances for clinical translation are in the spotlight, including biomarkers, molecular subtypes, and targeted therapies.