Improving diagnosis and monitoring of neonatal sepsis through machine learning analysis of clinical and transcriptomic data

Parkinson, Edward ORCID: https://orcid.org/0009-0006-4552-9495 2025. Improving diagnosis and monitoring of neonatal sepsis through machine learning analysis of clinical and transcriptomic data. PhD Thesis, Cardiff University. Item availability restricted.

	PDF (Edward Parkinson, PhD, Thesis) - Accepted Post-Print Version Restricted to Repository staff only until 7 November 2026 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (17MB)
	PDF (Cardiff University Electronic Publication Form) Restricted to Repository staff only Download (248kB)

Abstract

Neonatal sepsis, a dysregulated immune response to infection, presents a major global health challenge. Diagnosis and treatment monitoring remain difficult in this unique patient population, where blood cultures have high false negative rates, and adult diagnostic biomarkers show poor sensitivity and specificity. Consequently, many neonates receive unnecessary empirical antibiotics, with implications for side effects and antibiotic resistance. Recently, genomic biomarkers derived through statistical and machine learning (ML) analysis of gene expression profiles have demonstrated promising diagnostic accuracy. However, these gene signatures lack large-scale validation and may be overfit to small discovery cohorts. The potential of transcriptomic biomarkers for treatment monitoring also remains underexplored. To begin addressing large-scale validation using modern RNA-Sequencing data, we developed a novel sample quality assurance utility, and showed how independent gene filtering improves classification performance, feature selection stability, and biomarker interpretability. We examined biomarker generalisability by using gene module membership data to regularise penalised regression models. We developed a novel penalty weighting scheme that quantifies gene co-expression within a module and showed that favouring modules with higher co-expression yielded neonatal sepsis biomarkers with improved diagnostic performance, and biological relevance. Finally, we examined the host immune response to antibiotic treatment in neonates. We found that gene expression profiles characteristic of innate immune hyperinflammation and adaptive immune suppression in sepsis reverse early during treatment. Moreover, we used expression changes in treatment-responsive genes to define the immune module ratio (IMR), a molecular marker of immune response highly correlated with clinical outcomes and potentially valuable for prognosis. Genome-wide analysis also revealed a striking recovery phenotype characterised by reduced type 1 interferon signalling and increased antimicrobial gene expression. These results suggest that the dysregulated immune response is rapidly reversible, with implications for improved monitoring and targeted therapies.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Funders:	Cardiff University
Date of First Compliant Deposit:	7 November 2025
Date of Acceptance:	5 November 2025
Last Modified:	07 Nov 2025 11:02
URI:	https://orca.cardiff.ac.uk/id/eprint/182191

Actions (repository staff only)

Edit Item