A readily interpretable rule involving multiple forms of pairwise molecule comparisons with applications for clinical make-decision of breast cancer management.

Cells of breast cancer (BC) can metastasize to lymph nodes or other organs and is the high leading cause of female cancer deaths. Efforts to improve the performance of early detection and precise treatment are urgent and important for BC management. In clinical application, compared with highly complex classification functions, precise, simple and biologically interpretable algorithms can strengthen the understanding of disease development and facilitate the personalization of therapeutic strategies. In this study, a novel readily interpretable decision rule involving multiple forms of molecular relationship (RI-MFR) was proposed for cancer clinical management applications. In RI-MFR, linear and nonlinear pairwise molecule comparisons were comprehensively analyzed by a joint probability mass function for the identification of top scoring pairs from high dimensional biomolecular data. Based on the selected few molecule pairs, accurate, readily interpretable decision rules were inferred to provide biological insight as to how classification was performed. RI-MFR can effectively eliminate the influence of sample variability caused by individual differences. RI-MFR was successfully employed to analyze changes in metabolic mechanisms during BC development based on genomics datasets and metabolic alterations before and after BC therapy using our metabolomic profiling. The experimental results indicated that genes and metabolites involving in the glycosphingolipid metabolism may be the crucial factors associated with BC development and contribute to the enhanced effectiveness of BC treatment. Compared with other algorithms, RI-MFR had the significantly better classification results with p-values < 0.05, which suggested it is a more useful tool to identify important bioinformation for clinical BC management.