Session Name: Liver: Large Data and Artificial Intelligence
Session Date & Time: None. Available on demand.
*Purpose: Blood-based biomarkers distinguishing acute-rejection (AR) from other causes of graft dysfunction (acute dysfunction non-rejection (ADNR)) would have significant clinical applicability in liver transplant (LT) management. However, the high dimensionality of omics data, scarcity of clinical samples and individual diversity in gene expression levels decreases our power to detect disease-specific biomarkers. A logical first step to address these aforesaid issues is performing dimension-reduction and projection of data into a lower dimension space (2D or 3D), reflecting natural clustering of same-class samples. In our work we developed a Linear Discriminant Analysis (LDA) based classifier to differentiate AR and ADNR, using blood gene expression levels in LT patients.
*Methods: A commonly used unsupervised technique (PCA), only works when interclass variance is higher than intraclass variance. In our work we used a supervised technique (LDA), which projects data into a lower dimension space and facilitates grouping of same-category data points. We obtained the gene expression data of 91LT recipients (43 AR, 48 ADNR), from two previously published studies (CTOT14 and Northwestern University biorepository); and a symmetrical uncertainty filter was applied prior to LDA to remove highly correlated probes that destabilize the model. Transcriptome profiles based on whole blood samples were time-matched with biopsy results, whose phenotypes were directly determined from these studies.
*Results: We found that supervised methods outperformed unsupervised techniques in our heterogeneous human clinical data. Subsequently, we developed an LDA based classifier to classify AR and ADNR cases in LT recipients with graft dysfunction. The classifier had overall accuracy in the training set of 83.5% (76/91) in leave-one-out cross-validation. Sensitivity was 93.0% (40/43) and specificity was 75.0% (36/48); PPV was 76.9% (40/52), and NPV was 92.4% (36/39).
*Conclusions: Our work resulted in a highly accurate classifier to distinguish AR and ADNR; with clinical applications in allowing for non-invasive detection and treatment of AR with higher certainty and safety than current clinical tools.
To cite this abstract in AMA style:Miller M, Sinha R, Weems J, Holman J, Altrich M, Kleiboeker S, Levitsky J. Application of Linear Discriminant Analysis (lda) to Differentiate Acute Rejection (ar) and Acute Dysfunction Non-Rejection (adnr) in Liver Transplant Recipients [abstract]. Am J Transplant. 2021; 21 (suppl 3). https://atcmeetingabstracts.com/abstract/application-of-linear-discriminant-analysis-lda-to-differentiate-acute-rejection-ar-and-acute-dysfunction-non-rejection-adnr-in-liver-transplant-recipients/. Accessed June 13, 2021.
« Back to 2021 American Transplant Congress