COVID-19 patients show significant clinical heterogeneity in presentation and outcomes that makes pandemic control and strategy difficult; optimising management requires a systems biology approach of understanding the disease. Here we sought to understand and infer complex system-wide changes in patients infected with coronaviruses (SARS-CoV and SARS-CoV-2; n=38 and 57 samples) at two different disease stages compared with healthy individuals (n=16) and patients with other infections (n=144). We applied inferential statistics/machine-learning approaches (the COVID-engine platform) to RNA profiles derived from peripheral blood mononuclear cells (PBMCs). Compared to healthy individuals, an integrated blood-based gene signatures distinguished acute-like (mimicking coronavirus-infected patients with prolonged hospitalisation) from recovering-like patients. These signatures also hierarchically represented systems-level parameters associated with PBMC including dysregulated cytokines, genes, pathways, networks of pathways/concepts, immune status, and cell types. Proof-of-principle confirmatory observations included PBMC-associated increases in ACE2, cytokine storm-associated IL6, enhanced innate immunity (macrophages and neutrophils), and lower adaptive T and B cell immunity in patients with acute-like disease compared to those with recovery-like disease. Patients in the recovery-like stage had significantly enhanced TNF, IFN-g, anti-viral, HLA-DQA1, and HLA-F gene expression and cytolytic activity, and reduced pro-viral gene expression compared to those in the acute-like stage in PBMC. Besides, PBMC-derived surrogate-based approach revealed overlapping genes associated with comorbidities (associated diabetes), and disease-like symptoms (associated with thromboembolism, pneumonia, lung disease and septicaemia). Overall, our study involving PBMC-based RNA profiling may further help understand complex and variable systems-wide responses displayed by coronavirus-infected patients.