Gene expression profiling and machine learning to understand and predict primary graft dysfunction
2007 IEEE 7th International Symposium on BioInformatics and BioEngineering
Lung transplantation is the method of choice for the treatment of end-stage pulmonary diseases. A limited donor supply has dramatically increased the waiting time for transplant recipients. Approximately 4000 patients are currently on the transplant waiting list. Unfortunately, up to 10-20% of these patients will die from their underlying lung disease before an organ becomes available. Currently, only 10-20% of cadaveric donor organs offered for transplantation are judged to be acceptable under
... be acceptable under the current selection criteria. Of the donor lungs selected for transplantation, 15-30% of them fail due to primary graft dysfunction (PGD). PGD is a severe allograft ischemia-reperfusion (I/R) injury syndrome occurring in the hours following transplantation. It significantly affects morbidity as well as early and late mortality. This has resulted in an intense pressure to search for alternative selection criteria for selecting suitable donor lungs. In this study, we attempt to further our understanding of the gene products involved in PGD by observing the changes in gene expression across donor lungs that developed PGD versus those that did not. Our second goal is to use a machine learning technique -support vector machine, to distinguish donor lungs suitable for transplantation versus those that are not, based on the gene expression data. Results from microarray analysis produced a set of differentially expressed transcripts that were involved in signalling and apoptosis pathways. Various transcripts particular to stress-sensitive pathways were also identified. Results also indicate that the metallothionein gene, specifically metallothionein 3, may protect donor lungs from developing PGD. A classification accuracy of 70% was achieved, when a set of 100 differentially expressed transcripts was used to differentiate unsuitable donor lungs from suitable ones. This is the first such attempt to combine the identification of a molecular signature for PGD, using human samples, with machine learning methods for class (donor lung) prediction.