Predicting long-term kidney graft failure using novel multi-omic blood-based biomarkers and artificial intelligence tools.

Kidney transplantation (KT) remains the preferred treatment for end-stage renal disease. With advancements in immunosuppressive regimens and KT surveillance, graft survival has improved, though mainly in short-term. Meanwhile, aging populations with multimorbidity and expanding donor criteria shape a new landscape for KT management. Numerous prediction tools, including genomic, transcriptomic and/or proteomic panels or biomarkers, have been developed for short-to-interim outcomes, yet variable outcome definitions, modest samples and limited external replication preclude clinical utility. The temporal nature of association strength for graft failure risk factors reflects changes in underlying pathomechanisms and underscores the need for extensive validation. Chronic allograft rejection is a progressive process intertwined with variable T cell and antibody-mediated rejection patterns. On a molecular level, both innate and adaptive immune cells interface within the local graft microenvironment and release donor cell products (eg, exosomes, peptides, apoptotic bodies) that prime both T and B cell, but also IFNγ driven NK cell-mediated responses. Complement and Ig deposits along capillary lining lead to activated endothelium that promotes immune cell influx and aberrant differentiation patterns. Under cytokine and growth factor stimulation, mesenchymal transition of graft epithelial cells leads to altered extracellular turnover and TGFβ-mediated fibrosis. These mechanistic processes remain incompletely understood but represent a biologically plausible source for urine/blood biomarkers and omic profiling. Artifical intelligence and machine-learning tools provides a promise for elucidating the nature of these mechanisms due to their ability to capture non-linear trends and complex interactions. However, early efforts still remain unsatisfactory as the data demand increases, with concomitant requirements for high feature quality and sample representativeness.