Despite the improvement in global health over the last three decades, infectious diseases are still a major cause of morbidity and mortality globally, with the highest burden in developing countries and in children under 5 years. The heterogeneity in the clinical presentation of severely ill patients and the lack of rapid diagnostic and prognostic tools for the aetiological distinction of infecting pathogens complicates care decisions, leading to the indiscriminate use of antibiotics and consequently, increased antimicrobial resistance and mortality. Reducing childhood mortality and morbidity due to infectious diseases requires better diagnostics and prognostics, particularly in low-resource settings. Understanding the molecular processes that underlie different aetiologies and survival outcomes would enable the initiation of appropriate and timely treatment.

This thesis aimed to characterise protein and transcriptomic signatures associated with (i) different microbial aetiologies of severe acute infection and (ii) an elevated risk of death in African children hospitalised with severe acute infections. An untargeted high-performance liquid chromatography-tandem mass spectrometry (HPLC-MS/MS) approach was used to characterise the plasma proteome of children admitted to hospital with different infectious aetiologies. The resulting data was analysed using a collection of machine learning algorithms. The data was used to identify a protein signature with the highest power to discriminate between children with bacterial or viral infections. A novel protein microarray chip was then developed to validate the discovered signature in an independent cohort of hospitalised children with severe acute infections. Lastly, an RNA seq-based transcriptomics approach was used to characterise gene expression changes associated with post-admission outcomes. Using gene set enrichment and modular analysis, the correlation between gene expression and impending inpatient mortality was characterised.

Key findings from this work include a validated protein signature that could classify children according to bacterial or viral aetiology and a description of the immune response to severe disease that could be correlated with an increased likelihood of death in children admitted to hospital with severe acute infection. In the proteomics analysis, a random forest-derived protein signature made up of CRP, LBP, AGT, SERPINA1, SERPINA3, PON1 and HRG was identified that could correctly classify bacterial infection from viral infection in a held-out test set with AUC of 0.84 (95%CI 0.72 – 0.95). The performance of individual proteins in the signature was assessed in an independent cohort of children using a novel protein microarray chip and AGT had the best discriminatory power of 0.7 (95%CI 0.64 – 0.75) relative to a clinically-approved CRP test whose AUC was 0.76 (95%CI 0.68 – 0.84). In addition, there was an association between the expression levels of AGT, SERPINA1, PON1 and HRG with mortality. AGT was also associated with severe acute malnutrition in children. Functional analysis of the proteomics data showed that children with bacterial infections had an enrichment of acute phase responses and neutrophil degranulation pathways while platelet degranulation was negatively associated with bacterial infections.

Analysis of the transcriptomics data showed that imminent inpatient death was marked by a down regulation of CD8 T cell activation, type I interferon signalling and an over expression of the unfolded protein response and heme metabolism pathways.