A blueprint in bioinformatics: From transcriptomics to outcome prediction in kidney transplantation

Kidney transplantation is the preferred treatment for patients with end-stage kidney disease, yet allograft rejection remains a major cause of graft loss and morbidity. Accurate classification of rejection subtypes is essential to guide treatment strategies. While histopathological assessment has long been considered the gold standard in diagnosis, it is limited by interobserver variability. Molecular classification based on transcriptomic data has recently emerged as a promising complementary approach.
This thesis aimed to explore computational and machine learning methods for transcriptomic data and their application in characterising molecular differences among kidney allograft rejection subtypes. The first part focused on evaluating normalisation strategies for Bruker® nCounter data generated using the Banff-Human Organ Transplant (B-HOT) panel. RUVSeq was identified as the most effective method for minimising technical variation while retaining biologically relevant variation in multi-centre studies. Furthermore, we developed and validated a molecular classifier for distinguishing antibody-mediated rejection (AMR), T-cell-mediated rejection (TCMR), and non-rejection using publicly available microarray datasets. This approach demonstrated the feasibility of developing molecular diagnostics without the need for centralised data generation and identified six novel candidate genes for the panel.
The second part applied molecular profiling to clinical cohorts with distinct rejection subtypes. We found no direct associations between mast cell abundance and fibrosis across rejection subtypes, challenging previous assumptions about their role in tissue scarring. In addition, we showed that borderline TCMR (bTCMR) cases lacked a molecular profile consistent with acute TCMR, highlighting current diagnostic limitations for ambiguous subtypes. Finally, we revealed that chronic active AMR (CA-AMR) cases are molecularly distinct from patients presenting with transplant glomerulopathy and microvascular inflammation (cg+MVI) without C4d or donor-specific antibodies.
Overall, this work underscores the value of transcriptomic profiling in refining rejection subtype classification and emphasises the need for improved molecular diagnostics in kidney transplantation.