Molecular and histologic changes in graft rejection are not completely specific, often being induced by injuries and other diseases. We examined various methods for using molecules to predict rejection: molecular gene set means, single molecular classifiers derived by machine learning, and multiple classifiers.

We derived top 20 transcripts associated with (by t-test) histologic antibody-mediated rejection (ABMR) comparing biopsies with ABMR to biopsies with no rejection – everything else (EE) but including T cell-mediated rejection (TCMR) in the comparator, the AvsEET algorithm. The same algorithm was used to build a molecular classifier of ABMR, which used 20 transcripts selected by machine learning to obtain a classifier score.

Next, we compared the classifier scores to the gene set means. Logistic regression was used to predict the presence/absence of disease in a training set and then applied to the test set. AvEET classifier had better area under the curve (AUC) than the ABMR transcript set mean (0.84 vs. 0.80, respectively) (Figure 1 upper panel). To find if we can get better performance of AvsEET classifier, we built additional all rejection (ABMR, TCMR, Mixed) classifier (RvsEE) and TCMR classifier with ABMR in the comparator (TvsEEA) then combined them with the AvsEE classifier and again tested the prediction of histologic diagnosis of ABMR.

Combined classifiers improved the ability to predict histologic ABMR (AUC 0.86) compared to using the ABMR classifier alone, as demonstrated by the Net Reclassification Index (Figure 1, lower panel).

Thus the AvsEE molecular classifier is superior to the AvsEE transcript set mean in predicting histologic ABMR diagnoses, but the combination of three classifiers (RvEE, TvEEA, and AvEET) is superior to either a single classifier or to a transcript set means for predicting ABMR.

CITATION INFORMATION: Famulski K, Halloran P. Molecular Classifier of Antibody-Mediated Rejection Is a Better Predictor of Histologic Rejection Than a Corresponding Transcript Set and Is Further Improved When Combined with Classifiers of Rejection Syndromes. Am J Transplant. 2017;17 (suppl 3).

« Back to 2017 American Transplant Congress