Inference of protein-protein networks for triple-negative breast cancer using single-patient proteomic data

The advances in proteomic technologies have offered an unprecedented opportunity and valuable resources to reveal molecular targets for treatment. Although a number of approaches have been designed to develop mathematical models using the time series proteomic profiles, the recently published single-patient proteomic data raised substantial challenges for analysing these non-time series datasets. To address this issue, this work proposes the first attempt for designing mathematical models using the non-time series proteomic data. Using the triple-negative breast cancer (TNBC) as the test system, we first use the single-cell analysis algorithm to derive the pseudo-time trajectory of the protein activities. Our integrated approach includes both a top-down approach (namely the Gaussian graphical model) and a bottom-up approach (i.e. differential equation model) to reverse-engineer the regulatory network. Based on the information from GO-enrichment analysis and KEGG database, we select 16 proteins that are key components in the mitogen-activated protein (MAP) kinase pathways. We construct the structure of a network with 16 proteins and a dynamic model for a network of 12 proteins. The derived protein-protein relationships are partially supported by the established protein activation relationships, and our model also predicts potential protein relationships that may be confirmed by further experimental studies. In summary, our results suggest that the proposed integrated framework is an effective approach to reconstruct regulatory networks using non-time course proteomic data.